Citation
Efficient flow and congestion control for self-similar traffic in ATM networks

Material Information

Title:
Efficient flow and congestion control for self-similar traffic in ATM networks
Creator:
Fu, Wen-Yen, 1965-
Publication Date:
Language:
English
Physical Description:
x, 121 leaves : ill. ; 29 cm.

Subjects

Subjects / Keywords:
ATMs ( jstor )
Bandwidth ( jstor )
Feedback control ( jstor )
Local area networks ( jstor )
Simulations ( jstor )
Stochastic processes ( jstor )
Traffic characteristics ( jstor )
Traffic congestion ( jstor )
Traffic delay ( jstor )
Traffic models ( jstor )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 1998.
Bibliography:
Includes bibliographical references (leaves 116-120).
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Wen-Yen Fu.

Record Information

Source Institution:
University of Florida
Rights Management:
The University of Florida George A. Smathers Libraries respect the intellectual property rights of others and do not claim any copyright interest in this item. This item may be protected by copyright but is made available here under a claim of fair use (17 U.S.C. §107) for non-profit research and educational purposes. Users of this work have responsibility for determining copyright status prior to reusing, publishing or reproducing this item for purposes other than what is allowed by fair use or other copyright exemptions. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder. The Smathers Libraries would like to learn more about this item and invite individuals or organizations to contact the RDS coordinator (ufdissertations@uflib.ufl.edu) with any additional information they can provide.
Resource Identifier:
39546794 ( OCLC )
ocm39546794

Downloads

This item has the following downloads:


Full Text











EFFICIENT FLOW AND CONGESTION CONTROL FOR
SELF-SIMILAR TRAFFIC IN
ATM NETWORKS












BY

WEN-YEN FU











A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY


UNIVERSITY OF FLORIDA


1998
















ACKNOWLEDGEMENTS


I would like to thank the members of my supervisory committee for their help and guidance throughout this work. I would like to express special thanks to my committee chairman, Dr. Latchman, for his invaluable instruction and advice through this research. Also, special thanks to Dr. Chow, Dr. Couch, Dr. Taylor and Dr. Newman-Wolfe for their helpful suggestions and comments on this dissertation.

I am grateful to the many colleagues and friends who encouraged me to go on for my graduate career. My special thanks go to Dr. Shang-Yi (Debbie) Lu, whose advice and expertise have been invaluable to this work. I am indebted to my wife, Hui-hsien, and my son, Kevin, for accompanying me through the years without any complaint. Finally, I would like to thank my parents and my sisters for their full support and understanding during my long and fruitful education.






















ii
















TABLE OF CONTENTS




ACKNOWLEDGEMENTS ................... ......... ii

LIST OF TABLES ................................. vi

LIST OF FIGURES ................................ vii

ABSTRACT .................................... x

CHAPTERS

1 INTRODUCTION ............................ 1

1.1 Motivations and Objectives . ...................... 1
1.2 Dissertation Outline ......... ............... 3

2 BACKGROUND............................ 6

2.1 Asynchronous Transfer Mode .......... ......... 6
2.1.1 ATM Layered Structure ........ ... ....... 7
2.1.2 Physical Layer ....... .... ............. 8
2.1.3 ATM Layer .. ........... ................ 9
2.1.4 ATM Adaptation Layer ........ .......... 11
2.2 Quality of Service ................... ...... 14
2.2.1 Accuracy ......... ....... .. ......... 15
2.2.2 Speed .......... . ................ 16
2.3 Traffic Characteristics ................... ... 16
2.3.1 CBR Traffic ....... ........ ......... 17
2.3.2 VBR Traffic ......... ............... 17
2.3.3 ABR Traffic .............. .. .. .... 18
2.4 CPS100 Example of ATM Switching Design ....... .... 19
2.4.1 Interface Modules ................ . ... 20
2.4.2 Advantages and Limitations of CPS-100 Switch ..... 24

3 TRAFFIC MODELLING AND SELF-SIMILAR PROCESSES .... 26

3.1 Poisson-Based Traffic Models .... ........... .... 27
3.1.1 Poisson Process Models ................ ... 27
3.1.2 Markov Process Models .................. 28
3.1.3 Markov-Modulated Process Models ............ 29
3.2 Self-similar Processes ............. ...... ... 30
3.2.1 D efinitions . . . . . . . . . . . . . 32
3.2.2 Self-similarity and Long-Range Dependence ........ 34


iii









3.2.3 Self-similarity and Hurst Effect . . . . . . 36
3.2.4 Self-similarity and Slowly Decaying Variances . . . 36
3.3 Modeling of Self-Similarity ................... 37
3.3.1 Fractional Brownian Motion . . . . 37
3.3.2 Fractional Gaussian Noise ........ ........ . 38
3.3.3 Fractional ARIMA(p,d,q) Processes .......... .. 39
3.3.4 Self-Similarity Through Aggregation .......... .. 41
3.4 Self-similar Traffic Trace Generators ......... .... .. 42
3.4.1 Random Midpoint Displacement Algorithm ....... 42 3.4.2 Aggregation of Renewal Processes ........ . . 44
3.4.3 M/G/oo with Heavy-tailed Distributed Service Time . 44 3.4.4 ARIMA............... .. .. ...... 44
3.4.5 Approximation of Power Spectrum ............ 45

4 CONGESTION AND FLOW CONTROL .......... ..... 46

4.1 General Framework for Congestion Management and Control . 46 4.2 Flow Control Schemes with Feedback ......... ..... 49
4.2.1 Credit-Based Scheme .................. 49
4.2.2 Rate-Based Scheme .. ................. . 50
4.3 Impropriety of Poisson-Based Traffic Models .......... 52
4.3.1 Network Configuration ......... ..... ... 52
4.3.2 Effective Throughput Comparisons ......... ... 53
4.4 Classical Control Theory and Congestion Control .. .. ...... 55

5 PREVENTIVE FLOW CONTROL SCHEME FOR SELF-SIMILAR TRAFFIC ...... ........................... ...... 58

5.1 Introduction .......... ................... 59
5.2 Peak Rate Allocation ..... ........ ......... 60
5.3 Statistical Allocation ...... ................. 61
5.4 Cell Loss Probability in Self-similar Queuing Models ...... 61 5.5 Proposed Call Admission Control Algorithm .......... 67
5.6 Numerical Results and Discussions ................ 67
5.7 Conclusion .......... ................. 72

6 PID FLOW CONTROL SCHEME FOR VBR TRAFFIC ...... 73

6.1 Introduction .............. ..... ....... 74
6.2 Generic Video Transmission System with Feedback ...... . 77 6.3 Proposed Flow Control Scheme ............ ....... 79
6.4 Network Scenario for Simulations .. . .... ... .. .... 82
6.5 Numerical Results and Discussions .... ...... ... .... 83
6.6 Summary .............. ................ 89

7 INTEGRATED SERVICES PROTOCOLS AND TRAFFIC ANALYSIS 90

7.1 Protocols for Integrated Services ................. 91
7.1.1 Resource Reservation Protocol .............. 91
7.1.2 Real Time Transport Protocol. ........ . 93
7.1.3 IP Multicast Protocol and Multicast Backbone ...... 94
7.2 Traffic Measurements and Analysis ........... . . . 96
7.2.1 Network topology and Network tool ........... 96
7.2.2 Traffic Measurements and Analysis ............ 100


iv









8 CONCLUSION ................... ........... 106

8.1 ATM Networks and Self-similarity .. ........ .... .. 106
8.2 Preventive Flow Control ................ ...... 107
8.3 Feedback Flow Control ................... .... 107
8.4 Future Work ....... .. ... ........ 108

APPENDICES

A SCRIPTS FOR GENERATING SELF-SIMILAR TRAFFIC .... 110

A.1 Random Mid-point Displacement Algorithm ........ ... 110
A.2 Power Spectrum Density Approximation ............. 111

B SCRIPTS FOR ESTIMATING HURST PARAMETER ........ 113

B.1 Variance Time Plot Algorithm .................. 113
B.2 Rescaled Adjustment Algorithm .................. 114

REFERENCES ........... ... ....... ........... 116

BIOGRAPHICAL SKETCH ................... ........ 121


































V















LIST OF TABLES




2.1 Technology Comparison ........ ................. 7

2.2 Service Characteristics of AAL Layers .... ......... .... 14

2.3 QoS Parameters ........ ... ............. 15


4.1 Performance comparison of Poisson process and SSP traffic models. 55 5.1 Robustness of the proposed CAC scheme .............. .. 69

5.2 Guarantee of QoS ............. ............... 71


6.1 CLR and normalized utilization rate under various (KI, KD) ... . 85 6.2 CLR and normalized utilization rate vs. number of VBR VCs . .. 89 7.1 Arrival rate statistics ........ ............. .... 102
























vi
















LIST OF FIGURES




2.1 B-ISDN Protocol reference model ................... 8

2.2 ATM Cell Format at UNI and NNI .................. 10

2.3 TV Comercial ................... ..... ...... 18

2.4 TV News ............ .... .............. 19

2.5 Functional model of the CPS-100 switch ............ ... 21

2.6 Adapter Interface of a Shared Medium/Output Buffering ATM Switch 22 3.1 2-state Markov chain ........ ............ .... 29

3.2 Self-similarity of WWW traffic ... ........ .... .... 31

3.3 Autocorrelation function and long-range dependence ..... .... . 35

3.4 Fractional Brownian motion and fractional Gaussian noise ...... 43 4.1 Network scenario for Poisson vs. Self-similar traffic models ...... 53 4.2 Variance-time plot for synthesized sample paths ........... 54

4.3 Performance comparison of different flow control schemes ...... 55 4.4 Performance comparison of rate-based scheme I . .......... 56

4.5 Performance comparison of rate-based scheme II ....... .... 56

4.6 Generic Feedback Control System .......... .......... 57


5.1 Cell loss probability vs. buffer size .. ............. .. . 68

5.2 Estimated CLP vs. genuine Hurst parameters ............ 70

5.3 Number of admitted connections vs. buffer size ........... 71

5.4 Number of admitted connections vs. cell loss probability .... ... 72


vii









6.1 MPEG frame size and its probability density function ........ 76 6.2 Generic Video Transmission System with Feedback ......... 78

6.3 Generic tracking system and its application in networks .... ... 80 6.4 PID flow controller ......... .......... ... .... . 80

6.5 Network Scenario . .......... ............. ... 83

6.6 Functional model of CPE ...... ..... ......... ... 83

6.7 Transmitted and received frame size with no feedback control . 85 6.8 Cumulative number of cells sent and lost with no feedback control . 86 6.9 Transmitted and received frame size with feedback control ...... 87 6.10 Cumulative number of cells sent and lost with feedback control . . 88


7.1 RSVP Functioning Block Diagrams .................. 92

7.2 Reservation Merge of RSVP ................... ... 93

7.3 Multicat Backbone (MBone) ................... .... 95

7.4 Session directory (SDR) .... .... ....... ..... .. 97

7.5 Multicast session with audio/video streams ............. 98

7.6 Video and audio streams in a MBone session ........ ...... 99

7.7 Network applications and TCP/IP protocol stack ....... ... 99

7.8 LISTNET Topology ................. ....... .. 100

7.9 Traces in LANs and WANs .... ................ . 101

7.10 Variance -time plot ........ ................ . 102

7.11 QQ-plot for Mbone traffic ................... ..... 103

7.12 QQ-plot for TCP traffic ................... .... .. 104

7.13 QQ-plot for UDP traffic ................... ...... 104

7.14 QQ-plot for overall traffic ................... ..... 105








viii















Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy




EFFICIENT FLOW AND CONGESTION CONTROL FOR
SELF-SIMILAR TRAFFIC
IN
ATM NETWORKS

By

Wen-Yen Fu

May 1998


Chairman: Dr. Haniph A. Latchman Major Department: Electrical and Computer Engineering

Asynchronous Transfer Mode (ATM) has emerged as one of the most promising solutions for the next generation networks since being adopted as the transfer mode of broadband integrated services digital networks (B-ISDN). With the advent of the high speed, multimedia networks, ATM must support applications with diverse traffic characteristics and quality of service (QoS) requirements. An accurate traffic characterization and modeling is, therefore, of primary importance in implementing effective flow control strategies and in making efficient use of network resources in ATM networks.

Recent studies have verified self-similar or fractal-like behaviors of the traffic over local and wide area networks, which implies that it is more appropriate to model this traffic using a self-similar process than using a traditional Poisson-based



ix









process. In this Ph.D. research, we investigate the properties of self-similar processes in great detail. We present a queueing model with a fractional Gaussian noise arrival process and deterministic service time. This queueing model is radically distinct from conventional queueing models whose arrival process does not take into account the self-similarity of incoming traffic. Our simulation results show that the latter leads to a substantial amount of inaccuracy in terms of network performance. Based on the queueing model, we propose a preventive flow control scheme suited for self-similar traffic using a Connection Admission Control (CAC). This CAC takes into account the Hurst parameter of traffic sources. The numerical results show that the proposed CAC ensures the QoS and achieves more admitted connections than the peak rate allocation CAC does.

In this research, we also propose a rate-based flow control scheme suited for real-time traffic. The main goal of the flow control scheme is to increase the network utilization with a margin of degradation in cell loss ratio. The proposed flow control scheme predicts the evolution of buffer occupancy over time using a linear predictor. We use a Proportional-plus-Integral-plus-Derivative (PID) controller to update the optimum sending rate at the transmitter dynamically. The adaptive policy attempts to keep the buffer occupancy for each virtual channel at a steady level, and the simulation results show that the proposed scheme works effectively against network congestion.
















x














CHAPTER 1
INTRODUCTION

Asynchronous Transfer Mode (ATM) is one of the most promising and widely recognized solutions for the high speed, multimedia networks of next generation. The anticipation is mainly due to the scalable building blocks ATM provides to serve as a stable foundation for evolving public and private networks. ATM is capable of handling many kinds of information including voice, data, image and video in an integrated manner and was accepted as the transfer mode for Broadband Integrated Services Digital Networks (B-ISDN) by the International Telecommunications Union (ITU) in 1988. With the advent of B-ISDN, ATM must support applications with diverse traffic characteristics and quality of service (QoS) requirements. An accurate traffic characterization and modeling is, therefore, of primary importance in implementing effective flow control strategies and in making efficient use of network resources in ATM networks.

1.1 Motivations and Objectives

ATM has been effectively making inroads in the world-wide networks and has made substantial progress in developing key standards regarding signaling, traffic management and LAN emulation. However, there still exist some challenges that we need to address in order to facilitate the design and optimal performance of functional ATM networks. This Ph.D. research presents two of these key issues, namely an appropriate characterization and modeling of ATM traffic and effective congestion management schemes suitable for the self-similar or fractal-like traffic expected on ATM networks.



1






2


Recent studies based on the trace of local area network (LAN) traffic, mainly at Bellcore [1], have led to the conclusion that LAN traffic can not be adequately represented by traditional Markov-based models, but instead can be more appropriately matched by self-similar or fractal-like models. More recently, variable-bit-rate (VBR) video traffic [2, 3, 4] and wide area network (WAN) traffic [5] was also found to exhibit self-similar characteristics. With the foreseeable prevalence of Internet services and multimedia applications, ATM networks will be carrying a substantial portion of self-similar traffic.

As we know when a number of bursty traffic sources are active and no interactive or preventive action is taken to against it, cell losses tend to occur due to buffer overflows. Despite that researchers have proposed many flow and congestion control schemes to solve the problems, whether these approaches take advantage of techniques in feedback control [6, 7, 8], call admission control (CAC) [9], traffic shaping and smoothing [10], or resource reservation and buffer management [11, 12, 13, 14], none of them takes the self-similarity of the traffic into consideration. As a consequence, a new flow control mechanism incorporating more realistic traffic models is necessary to avoid the tending degradation [15] of network performance and, at the same time to maintain better network utilization.

The main objective of this research is to design an effective flow and congestion control scheme for ATM networks by developing a queueing model to gain better understandings of the behaviors of a network whose incoming traffic is self-similar in nature. The primary goal of the flow and congestion control mechanism is to minimize the impact of traffic overload which induces congestion as well as cell loss, and, simultaneously, to achieve a better network utilization. In addition, this research will also undertake a tremendous efforts to develop a hierarchically structured testbed for exploring network performance and various flow control schemes for ATM networks under mixed self-similar and Poisson traffic.






3


1.2 Dissertation Outline

This dissertation is organized as follows: In Chapter 2, we give an introduction of ATM technology along with various traffic characteristics in ATM networks. We describe the ATM cell structure in great detail. The bottom three layers of B-ISDN reference model, namely Physical Layer, ATM Transport Layer and ATM Adaptation Layer are presented. Traffic characteristics of Constant Bit-rate (CBR), Variable Bitrate (VBR) and Available Bit-rate (ABR) or Best-effort services and QoS in ATM networks are discussed at the end of this chapter.

In Chapter 3, we first review traditional traffic models including Poisson, Markov and Markov-Modulated Poisson model. We then focus on self-similar processes and their properties and implication for novel self-similar traffic models. Two of the most important properties of self-similar processes, namely, Hurst effect and long range dependence will be discussed and analyzed completely. We introduce variancetime plots (VTP) and rescaled adjusted (R/S) statistics to estimate the Hurst parameter of a self-similar sample path. We present several synthesized algorithms for generating self-similar samples, for example, Random Mid-point Displacement (RMD) algorithm and FFT approximation. In this research, most algorithms for estimating the Hurst parameter and generating self-similar sample path are converted in Matlab scripts.

In Chapter 4, we introduce a traffic control and management framework in ATM networks. Important parts of the framework are Connection Admission Control (CAC), Resource Management (RM), Traffic Shaping, Feedback Control, and so on. We describe several examples including rate-basted and credit-based flow control schemes. The performance of several congestion control schemes are re-evaluated by using self-similar traffic models.






4


In Chapter 5, we propose a preventive flow control scheme using a Connection Admission Control (CAC) suited for generic self-similar traffic, particularly multimedia traffic in ATM networks. This CAC takes into account the Hurst parameter of incoming traffic sources. To facilitate the CAC, we derive the cell loss probability for a queueing system with an fGn arrival process and deterministic service time. The CAC is then implemented based on an upper bound of this cell loss probability. In addition, a number of numerical results are provided to validate the effectiveness of the proposed CAC scheme.

In Chapter 6, we propose an adaptive rate-based flow control scheme for realtime VBR traffic in ATM networks. The goal of the scheme is to minimize the impact of traffic overload in order to limit the cell loss rate to an acceptable range and also increase the network utilization. The proposed flow control scheme is based on predicting the evolution of buffer occupancy over time using a Proportional-plusIntegral-plus-Derivative (PID) controller and a linear predictor to adaptively update the optimum data emission rate at the transmitter. The adaptive policy attempts to keep the buffer occupancy for each virtual channel at a steady level and the simulation results show that the proposed scheme works effectively against network congestion. Along with the design of the new flow control scheme, we also develop a hierarchically structured testbed to measure network performance and explore various flow control schemes in ATM networks with diverse classes of incoming traffic.

In Chapter 7, we present three protocols which are playing key roles in the success of the future Internet Multimedia Network: Resource Reservation Protocol (RSVP), Real-time Transport Protocol/Real-time Transport Control Protocol (RTP/RTCP), and IP Multicasting Protocol-MBone. To study network dynamics and real traffic characteristics, we conducted traffic measurements and analysis based on empirical traces collected in the Laboratory for the Information Systems and Telecommunications (LIST) in August 1997. We illustrate traces standing for more






5


than 500,000 packets in different time scales. The traces are also treated separately according to the protocols of interest to us, which are IP Multicasting, Transport Control Protocol (TCP) and User Datagram Protocol (UDP). And finally, we conclude the dissertation with summary and future directions in Chapter 8.















CHAPTER 2
BACKGROUND

2.1 Asynchronous Transfer Mode

The ATM standard is primarily defined by International Telegraph and Telephone Consultative Committee (CCITT), lately called International Telecommunications Union-Telecommunication standardization sector (ITU-T). Some interim standards for some aspects of ATM have been developed by a user and vendor group known as ATM Forum in the absence of CCITT standards [16, 17, 18]. ATM as an emerging, cell-based technology is expected to integrate the currently separate networks used for voice, video, and data applications. The merger of these networks has the potential to provide us significant benefits and convenience. ATM, originally bonded to emerging Telecommunications technology standards, is an established way to provide channels with arbitrary bandwidth within a multiplexing hierarchy consisting of a well defined set of fixed bandwidth channels. Eventually, a side-effect of this provision of arbitrary-capacity channels, ATM will also be able to support channels of variable bit rate, and hence, will be capable of achieving a statistical multiplexing gain (SMG). This service is currently available in speeds ranging from 1.544 Mbps (TI or DS-1) to 155 Mbps (Optical Carrier Level 3 or OC-3), with rates expected to be available down to 64 Kbps and up to 622 Mbps (OC-12) soon. Ultimately, ATM is expected to scale up to the gigabit range.

As a promising candidate for networking in LANs as well as a replacement for Time Division Multiplexing (TDM) transmission systems in WANs, researchers routinely saw ATM as a scalable method for the provision of high-speed network





6






7

Attributes Datacom Telecom ATM
Traffic Type data voice data, voice, video
Switching packet circuit cell
Unit variable packet fixed frame fixed cell
Quality best effort guaranteed QoS
Access to Media shared dedicated dedicated
Connection connectionless connection-oriented connection-oriented Table 2.1: Technology Comparison

connections to routers and end systems. This trend has been so marked that it overtakes the development of ATM as a back-bone transmission method and displaces other development prospects for High-Speed LANs and WANs such as Frame Relay

(FR), Fiber Distributed Digital Interface (FDDI), and Switched Multimegabit Data Services (SMDS). Additionally, ATM is significantly different from both the popular data communication and telecommunication technologies in use today. ATM technology stands out by unifying these two worlds by borrowing desirable attributes from both data communication and telecommunication technologies [19]. The comparison between data communication, telecommunication, and ATM technology is shown in Table 2.1. Explicit information regarding ATM will be discussed in the next few sections.



2.1.1 ATM Layered Structure

The ATM protocol uses a layered structure, B-ISDN protocol reference model (see Figure 2.1), similar to the Open System Interconnect (OSI) model. However, the three-dimensional layer structure of ATM protocol is distinct from all other protocol models in the use of three planes across all three ATM layers plus a higher layer. The user plane is used for end-to-end data transfer, the control plane supports signalling used to established a connection, and the management plane supports management






8



"' Management Plane
a control Plani User Plane

Higher Layer tHigher Layer

AAL Layer

ATM Layer

Physical Layer



Figure 2.1: B-ISDN Protocol reference model and control functions for each network and its endpoints. The primary layers of the BISDN protocol reference model are the Physical Layer that provides for transmission of ATM cells over a physical medium that connects two ATM devices, the ATM Layer where cell segmentation and reassembly occurs, and ATM Adaptation Layer (AAL) that provides support for higher layer services such as circuit emulation, frame relay and SMDS.

2.1.2 Physical Layer

The Physical Layer of the ATM stack is divided into two sublayers: the Transmission Convergence (TC) sublayer and the Physical Medium (PM) sublayer. The TC sublayer is responsible for such tasks as line coding and bit timing, the generation and verification of the header error control byte, cell mapping and cell delineation and the PM sublayer provides bit-transmission capabilities on any particular medium including optical fiber, coaxial cable, Unshielded Twist Pair (UTP), and radio links.

Users have quiet a range of Physical Layers to choose from. The ATM Forum [20] has issued recommendations for 100 Mpbs and 140 Mbps transmission over multimode optical fiber, a 155 Mbps Synchronous Optical Network (SONET) interface, a 45 Mbps T3 interface, and is in the process of drawing up recommendations for






9


transmission over unshielded twisted pair (UTP). The CCITT/ITU-T has concerned itself mainly with standards suitable for use with SDH (Synchronous Digital Hierarchy) /SONET transmission networks and has defined standards for optical fiber at speed of 155 Mbps, 622 Mbps, and 2.4 Gbps. Some other relevant CCITT/ITU-T standards for the Physical Layer are G.703, G.707, G.708, G.709 and 1.432 [20, 16]. Proposals to transmit ATM cells at 2 Mbps, 1.5 Mbps, and other speeds also exist.

2.1.3 ATM Layer

The ATM layer is specified in ITU-T Recommendation 1.361 to provide for transparent transport of the data between AAL entities. This transfer takes place on a pre-established ATM connection according to a transfer contract consisting of a Quality of Service (QoS) class, a vector of traffic parameters. The ATM layer takes streams of 48-byte cell payloads as its input from higher layers, performs cell header generation, and passes cells to the TC sublayer of the physical layer such that order is preserved within virtual channels (VCs). Two levels of VCs can be supported at the user-network interface (UNI) that connects customer premises equipment (CPE) and ATM switches:


1. A point-to-point, or point-to-multipoint, Virtual Channel Connection (VCC)

consisting of a single connection established between two ATM VCC end points.

2. A point-to-point, or point-to-multipoint, Virtual Path Connection (VPC) consisting of a bundle of connections established between two ATM VPC end

points.


The ATM cell structure, in Figure 2.2, shows the first five bytes are allocated to the header of a fixed 53-byte cell. The entire header is protected by a 1-byte Header Error Check (FEC) field. We noticed that there is no retransmission of lost or corrupted data preformed by this layer. An access flow control mechanism may






10


| 1 byte 0 |I 1 byte a
I I I I I I I I
GFC VPI VPI
VPI VCI VPI VCI VCI VCI

VCI PT CLP VCI PT CLP

HEC HEC


PAYLOAD (48 bytes) PAYLOAD (48 bytes)


(a) at UNI (b) at NNI


Figure 2.2: ATM Cell Format at UNI and NNI

be implemented at the UNI, although currently no consensus exists as to how this will work. This access flow control will use the four Generic Flow Control (GFC) bits and is only meaningful when there is shared access to one ATM UNI port. At the receiving end, GFC receives as input a stream of cells from the Physical layer, from which it performs header extraction and delivers cell, in order, to the appropriate AAL service access point (SAP) using the Virtual Channel Identifier/Virtual Path Identifier (VCI/VPI) values as identifiers. At switching elements, the ATM layer uses the VCI/VPI to route the cells. While the VPI and VCI values may change at each switching element, the ATM layer does this translation.

Interpretation of the values in the Payload Types (PT) and Cell-Loss Priority (CLP) fields is done at the ATM layer in switching elements. The main feature of the PT Identifier is to discriminate user cells that carry user information from nonuser cells [20]. The PT Identifier of a user cell may not only indicate whether the cell has experienced congestion with Explicit Forward Congestion Indication (EFCI), but also suggest whether it contains an indication to the AAL protocol. For example, the






11


value "011" in the PT field suggests a user cell from AAL5 service class and having experienced congestion, and the value "110" simply indicates a Resource Management

(RM) cell.

The CLP field may be used for loss priority indication by the ATM endpoint and for selective cell discarding in network equipment. In a given ATM connection, and for each user-data cell in the connection, the ATM equipment that first emits the cell can set the CLP bit equal to 0 or 1. The CLP bit is used to distinguish between cells of an ATM connection: A CLP bit equal to 0 indicates a higher priority cell and a CLP bit equal to 1 indicates a lower priority cell which is subject to discard upon network congestion.

The UNI and Network/Network interface (NNI) are similar, the difference being that the UNI will connect access ports of ATM switches and Customer Premise Equipment (CPE) which could include broadband terminals, terminal adaptors, and cell-based LAN/WAN equipment, while the NNI can only connect trunk ports of ATM switches. The NNI is intended for ATM subnetworks or networks and, hence, does not need the GFC field in the ATM cell header. Moreover, the GFC field has been subsumed into the VPI field, allowing 16 times as many virtual paths. There exist a number of different UNI/NNI specifications, each of which is basically a "stack" comprising one particular physical layer and the ATM layer. In addition, the UNI/NNI encompasses signaling and connection management aspects from the control plane.



2.1.4 ATM Adaptation Layer

From previous sections, the ATM layer not only uses the service of the physical layer to transport cells, but also delivers cell payloads to the upper layers. In most cases, it is necessary to perform some adaptation functions, for instance, to






12


transform a sequence of cells into the byte stream used to transport digitized signals. This is performed by the ATM Adaptation Layer (AAL), as illustrated in Figure 2.1. The ATM adaptation layer (AAL) converts the real data stream into a cell stream and vice versa. Given the wide variety of possible characteristics of the input data streams, one would expect to find a variety of different adaptation layers. ITU Recommendation 1.362 defines the basic principles and classification of AAL functions, which are described as follows.


AAL Services Classes


Class A: constant bit-rate (CBR) service with end-to-end timing, connectionoriented. The examples are fixed-rate video and the circuit emulation services

such as DS1 or DS3 transport.

Class B: variable bit-rate (VBR) service with end-to-end timing, connectionoriented. The examples are packetized voice and video.

Class C: variable bit-rate (VBR) service with no end-to-end timing, connectionoriented

Class D: variable bit-rate (VBR) service with no end-to-end timing, connectionless


AAL Functions


AAL1 is designed to transport connection-oriented, CBR data stream in such

a way that clock information can be recovered at the receiving end (time transparency). AAL1 is, in effect, a virtual wire. Moreover, it can correct single-bit

errors in the payload and notifies lost cells or misordered cells.

AAL2 is specified to transport connection-oriented, VBR data streams in such

a way that timing information is recoverable at the receiving end. AAL2 has






13


become a key protocol in ATM implementation requiring support for VBR audio and video, for example, Motion Picture Expert Group (MPEG) video.

In theory, this AAL could be used to synchronize separate streams of data without the need for elaborate timestamping in the data streams themselves.

Due to the similar timing requirement as AAL1, AAL2 can also correct singlebit errors in the payload and notifies lost cells or misordered cells.

* AAL3/4 is the result of standards efforts for two AALs: AAL3 and AAL4,

which are both for transport of VBR streams without explicit timing information. AAL3 is connection-oriented and AAL4 is connectionless, although it is unclear what meaning this distinction actually has in ATM, and usually these two services are lumped together. As it turns out, the (then) CCITT merged AALs 3 and 4 into one single 3/4 type whose procedures can be applied in a connection-oriented or connectionless manner. Both AAL3 and AAL4 provide assured and non-assured services. In the assured mode, the AAL guarantees delivery of AAL Service Data Units (SDU) in order, and any lost SDUs are retransmitted. In the non-assured mode, this function is performed by higher layers. The main advantage AAL3/4 has is that it carries a length indication in its first cell, and, hence, make easier some traffic management function such

as fast buffer reservation.

* AAL5, also known as "Simple and Efficient Adaptation Layer (SEAL)," was

developed in response to a perception that AAL3/4 was inefficient, since only 44 octets of a cell payload carry actual data. AAL5 is a substantially lean AAL compared with AAL3/4 at the expense of error recovery and built-in retransmission. The trade-off provides a smaller bandwidth overhead, simpler processing requirements, and reduced implementation complexity. In AAL5, there is very little AAL overhead, with just one header and one 32-bit checksum






14

Service Class Class A Class B Class C ClassD
Delay Sensitive x x
Constant Bit Rate x
Variable Bit Rate x
Connection-oriented x x x x Connectionless x x
AAL AAL1 AAL2 AAL3/4 AAL5
Example DS1, DS3 MPEG Video SMDS TCP/IP

Table 2.2: Service Characteristics of AAL Layers

at the end of a frame which may comprise up to 1365 cells. It has been shown that AAL5 is at least as effective as AAL3/4 in detecting misordered or lost

cells.

Table 2.2 concludes the characteristics of all AAL service classes and their corresponding layers.





2.2 Quality of Service

ATM networks are expected to be one of the most dominant information infrastructures in the near future. However, many issues still need to be resolved before they can meet the glowing expectation. One major challenge, for example, is to design an efficient CAC and congestion control mechanism suitable for diverse traffic characteristics in the networks. To facilitate efficient and functional networks, comprehensive of the characteristics and QoS requirements of the traffic to be carried is of primary importance.

The primary objective of ATM is to provide the cell transfer in a connection with guaranteed performance. The performance parameters, or QoS classes, are summarized in Table 2.3. A QoS class with a connection can have either specified or unspecified performance parameters. A Specified QoS class specifies a set of performance parameters and the target value for the corresponding parameter. In the






15

Attribute Accuracy Speed
Cell Error Ratio x Cell Loss Rate x Cell Misinsertion Rate x Cell Transfer Delay x Cell Delay Variation x Table 2.3: QoS Parameters unspecified QoS class, no value is specified for the performance parameters, even though the network may assign those internal objectives. The performance parameters as defined in [20] can be further divided in two categories, namely accuracy and speed, as shown in Table 2.3.

2.2.1 Accuracy

The QoS classes regarding accuracy include Cell Error Ratio (CER), Cell Loss Rate (CLR) and Cell Misinsertion Rate (CMR). The CER for an ATM connection is defined as

Number of erroneous cells CER =
Total number of successful transmitted and erroneous cells '

where successful and erroneous cells found in severely faulty cells block are excluded. A cell block is a sequence of cells transmitted consecutively on a connection. A severely faulty cell block occurs when more than a certain number of error cells, lost cells or misinserted cells are observed in a received cell block.

The CLR is defined for an ATM connection (VCC or VPC) as Number of lost Cells CLR =
Number of transmitted Cells* It is noticed that lost and transmitted cells found in severely faulty blocks should be excluded from the cell population in computing CLR.

The CMR for an VCC or VPC connection is defined as Number of Misinserted Cells CMR =
Time interval






16


Again, misinserted cells appearing in severely faulty blocks should be excluded. Cell misinsertion on a particular connection is most likely caused by an undetected error in the header of a cell being transmitted on a different connection. This performance parameter is defined as a rate since the mechanism producing mis-inserted cells is independent of the number of transmitted cells received on the corresponding connection.

2.2.2 Speed

The QoS classes regarding speed include Cell Transfer Delay (CTD) and Cell Delay Variation (CDV). The CTD is defined as the elapsed time between a cell exit from the source and the corresponding cell at destination. The CDV is also called "jitter." There are two performance parameters associated with CDV: 1-point CDV and 2-point CDV. The 1-point CDV describes variability in the pattern of cell arrival events observed at a single measurement point with reference to a negotiated peak rate. The 2-point CDV describes variability in the pattern of cell arrival events observed at the output of a connection portion with reference to the pattern of the corresponding events observed at the input to the connection portion.

2.3 Traffic Characteristics

Network traffic can be characterized in many ways depending on the adapted criterion. One of the most widely used criterion is the tolerance of information delay or loss. For instance, real-time traffic, such as that for CBR service, is distinguished by its stringent CTD and CDV requirements, whereas best effort traffic, such as data transmission, demands superior CLR but yields inferior performance in terms of CTD and CDV. Consequently, an improper call admission control algorithm could not only underutilize network resources, but also make difficult the delivery of the QoS of eligible users. Another example for traffic characteristics is that some CBR traffic such as voice can tolerate a certain portion of cell discarding in a congested






17


situation and still remain within the range of good quality, but for some mission critical data service, even a single bit of data lost could result in the redundancy of the entire data received.

2.3.1 CBR Traffic

CBR traffic, generated by existing circuit switched systems as well as packetized voice and video sources using fixed rated coders, will be a dominant traffic in evolving broadband networks. Such traffic types are characterized by stringent delay requirements. Because of the periodic nature of the traffic, if a packet from a given source experiences a long delay (or is blocked due to the buffer overflow), then successive cells from the same source would experiences the same delay (or will be blocked) until the superposed arrival pattern undergoes a change.

2.3.2 VBR Traffic

The traffic generated by a typical source, in general, either alternates between active and silent periods and/or has a varying bit rate generated continuously. Furthermore, the peak-to-average bit rate of a VBR source is often much greater than one. Presenting VBR traffic to the network as CBR traffic by means of buffer smoothing has the drawback of underutilization of network resources and QoS degradation. Although doing so would simplify the network management task, it is more natural to provide VBR service to VBR sources to achieve higher resource utilization.

The characteristics of the video signal depend primarily on two factors, the nature of the video scene, and the type of VBR coding technique employed. Figure 2.3 and Figure 2.4 show the frame size over elapsed time of two typical MPEG II videos generated from recorded TV news and commercial. Several models were introduced to characterize video traffic with and without scene changes. For instance, a first order AR model was used to approximate the output rate of video sources. The






18


model provides an accurate approximation of the bit rate of a single video source without scene changes, but is not quite useful for queueing analysis.


x 104 TV commercial
6

5

4

3

2

1

0 20 40 60 80 100 frame


Figure 2.3: TV Comercial


2.3.3 ABR Traffic

Available bit-rate (ABR) traffic is usually referred to data applications, i.e., any application that is not voice, audio, video, or still image. Despite the fact that data networks have been operational for a number of decades, traffic characteristics of data sources are not well understood. The main difficulty comes from the fact that there does not exist any typical data connection. The problem of characterizing the source characteristics of ABR traffic is further complicated by the difficulty of predicting in advance the traffic characteristics of a connection, even if the particular application type is known. For example, in client-server computing, the amount of data exchanged and the source behavior may differ significantly from one application to another. Furthermore, data connections are not generally established between two users, but between a group of users, as in the case of local area network (LAN) interconnection.






19



x 10' TV News
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0 20 40 60 80 100 frame


Figure 2.4: TV News


The fast growth in the number of personal computers and workstations and the rising need for interconnecting them together with, for example, servers and printers in an office has led to the fast expansion of LANs. As the need for interoffice communication rose, LANs were in turn interconnected via bridges and routers. Consequently, users soon started to demand scalable throughput for large volumes of data exchange between them. In addition, emerging multimedia applications that integrate voice, data, image, and video started to change the face of networking in local and wide areas due to their real time and high bandwidth requirements.

2.4 CPS100 Example of ATM Switching Design

In spite of the strong competition from the evolving Gigabit Ethernet technology, more and more research and development have been involved with ATM switching technology over the years. The main reason is that ATM switches have proved to be the best option in the backbone of wide area networks (WANs). In this section, we present a fast packet switch architecture based on a shared medium and output buffering scheme for which simulation and performance studies have been conducted






20


at the University of Florida [21]. The CPS-100 switchI is an ATM-based switching system that supports ATM and various data services such as Switch Megabyte Data Service (SMDS) and Frame Relay for public networks. CPS-100 switch is a time division, shared medium switch fabric interconnecting time division, shared memory switching elements. The switch structure is generally composed of the following four major components:


1. The switching fabric which performs the basic switching function of transferring

cells or packets from input to output. The switching fabric in the CPS-100 switch is composed of a high-speed Virtual ATM (VATM) backplane bus and

is shared with all interface modules.

2. The access interface module provides an interface for customers to access the

services supported by network. The Customer Premises Equipment (CPE) attaches to an access interface module for transmitting and receiving data via

a dedicated link.

3. The trunk interface module provides an interface for interconnecting switching

systems to support inter-switch communications.

4. The CPU module communicates with all the interface modules in the switch

and performs the high-level control functions, such as connection establishment and release, memory administration, bandwidth allocation, and maintenance.


2.4.1 Interface Modules

Figure 2.5 illustrates the functional model of a six-port CPS-100 switch. Note that an access/trunk interface module can support multiple bidirectional links. Both external trunks and CPE links are connected to the VATM bus via the access/trunk

1The CPS-100 was developed by Loral Data Systems, Sarasota, FL.






21




Access To the Trunk CPE Interface Interface Module Module of other CPS 100 Switch
Trunk
CPE Interface Module
Access
CPE Interface
Module

CPE .............................................:





Figure 2.5: Functional model of the CPS-100 switch

interface module which consists principally of two controllers, namely an ingress adaptor and an egress adaptor. The ingress adaptor receives cells from the incoming link, buffers them as necessary, and sends them through the switching fabric. Similarly, the egress adaptor accepts cells from the switching fabric, provides any necessary buffering, and transmits the cells over the outgoing link. The breakdowns of the two adaptors are shown in Figure 2.6. These adaptors also perform all the cell-by-cell processing functions such as Virtual Connection Identifier (VCI) translation, priority assignment and routing. Since the VCI is generally only local to each switch port, the VCI of each cell must be translated to the value assigned for the succeeding links. This operation is performed by a VCI translation table.

The ingress adaptor appends routing overhead on an incoming cell to specify the output link associated with the virtual connection to which the cell belongs. The ingress adaptor also looks up other connection related information such as the loss priority and the service priority of the cell to assist the egress adaptor in making its buffering and service scheduling decisions. Since all the cells of a virtual connection






22











From Link A/B To Link A To Link B FIFO FIFO FIFO




Ingress Egress Egress Message Message Message Processor Processor Processor Egress Egress
Buffer I Buffer Memory Memory Egress
Protocol
Processor


FIFO a FIFO







Figure 2.6: Adapter Interface of a Shared Medium/Output Buffering ATM Switch






23


will follow the same route in traversing a switch, the sequence of cells is preserved without the need of explicit sequence numbers.

The most important function of the egress adaptor is to temporarily buffer those cells which cannot be delivered to a receiver immediately. A finite buffer memory allocated in the egress adaptor is for resolving output contention. If congestion causes the buffers to fill up, the controller has no choice but to discard cells. The buffer thus forms a critical resource that needs to be adequately controlled.

A modified Partial Buffer Sharing scheme is implemented in the egress adaptor to ensure satisfactory loss rates for different loss priority classes and fairness in sharing the resource among competing users. Moreover, a priority-dependent service scheduling scheme is realized to control the order in which cells are sent from the buffer to the output link. The service scheduling control is necessary for meeting the different switching delay requirements for different service priority classes. Service and loss priorities are assigned to each virtual connection during connection establishment based on the delay and loss requirements specified for the particular connection.

It is noticed that only one ingress adaptor can have access to the high-speed bus at any time. So a bus arbitration scheme is implemented for resolving contention for access among different ingress adaptors. Since the shared medium operates at a substantially higher speed (approximately 5 Gbps) than that of link interface modules, the CPS-100 can be viewed as a non-blocking switch, having no internal buffering and with no contention visible from outside the switching fabric. Though the high speed of the switching fabric reduces input contention, it also results in major queueing taking place at the buffer of egress adaptors. The proposed flow control scheme can be used along with existing buffer management schemes to avoid immense buffer overflows.






24


2.4.2 Advantages and Limitations of CPS-100 Switch

A principal advantage of the CPS-100 ATM switch structure is its flexibility to support multiple kinds of port interfaces. It has been widely recognized that ATM technology is targeted to support a rich mixture of broadband and narrowband services as well as the high-capacity interconnection of existing data networking applications. Thus, it is important for a switching node to provide different kinds of interfaces for the diverse applications in a highly flexible and expandable way. In the CPS-100 switch, this is achieved through a modular design that places the ports on separate interface modules. Various combinations of port interfaces, link access speeds and physical media can be supported easily by implementing different interface modules on the switch. A CPS-100 switch can be configured to support up to 16 of these interface modules.

Another advantage of the CPS-100 switch is that with this architecture dynamic priority functions and multicast operations can be supported flexibly with no additional hardware complexity. This function is preferable since some applications such as ATM LANs require the ability to support multiple guaranteed classes of services and full-bandwidth multicasting. Each interface module can implement suitable buffer management and scheduling policies based on the traffic characteristic as well as loading conditions to best serve the delay and loss requirements of different traffic classes. In addition, since the major queueing takes place at the output buffer, each interface module can perform load monitoring on the buffer usages of different traffic classes to support congestion control operation. Lastly, the buffer space within an interface module can be allocated dynamically to each port so as to achieve a lower probability of memory overflow.

On the other hand, as with other shared medium/output buffering ATM switches, the CPS-100 switch architecture has the typical limitation of requiring the






25


high speed switching fabric to accommodate simultaneous arrivals from different input ports. The switching fabric and output buffer must effectively operate at a much higher speed than that of each interface port. The implementation of the high-speed bus and buffer memory could be complex if the required memory access speed is very high, thus limiting the capability of the switch to support very high speed interface ports.















CHAPTER 3
TRAFFIC MODELLING AND SELF-SIMILAR PROCESSES

A recently observed phenomenon related to traffic in Broadband Integrated Service Digital Networks (B-ISDN) is the "self-similarity" or "burstiness" exhibited by key services such as compressed video [3], file transfer [5], and WWW [5]. Two main sources of burstiness are due to the shapes of the marginal distribution as well as the autocorrelation function of packet arrival rate or interarrival times [22].

When modeling network traffic, packet and connection arrivals are often assumed to be Poisson processes because such processes have attractive theoretical properties. A number of studies have shown, however, that for both local-area and wide-area network traffic, the distribution of packets interarrival time clearly differs from exponential [23, 24, 25, 26, 27]. Recent works argue convincingly that LAN traffic is much better modeled using statistically self-similar processes, which have significantly different theoretical properties than Poisson processes [28, 2]. An appropriate traffic modeling which reflects the burstiness or self-similarity is, hence, critical to the success of performance evaluation.

The remainder of this chapter is organized as follows. Commonly used Poisson and Poisson-based traffic models are presented in Section 1. The self-similar processes and their properties are given in Section 2. Section 3 describes the modeling of Selfsimilar processes, and finally several popular self-similar processes generators are described in Section 4.








26






27


3.1 Poisson-Based Traffic Models

3.1.1 Poisson Process Models

Poisson process models are the oldest traffic models, dating back to the advent of telephony primarily contributed by the well-known pioneering engineer A. K. Erlang in the 1910s. It is well known that the traffic on telecommunication networks can be modelled and characterized by a Poisson process. One of the most important traffic characteristics on telecommunication networks is that most human initiated processes are memoryless. A Poisson process can be characterized as a renewal process whose interarrival times {I,,} are exponentially distributed with rate of A, that is


Pr {I, < t} = 1 exp(-At). (3.1) Equivalently, it is a counting process, satisfying


Pr{A(t) = n) = (At)- exp(-At), (3.2) n!
where A(t) is the number of arrivals in a time interval (0, t).

Poisson processes enjoy some elegant analytical properties such as:


1. The number of arrivals in disjoint intervals are independent, which is referred

to as "independent increment" property.

2. Merging independent Poisson processes result in a new Poisson process with a

rate being the sum of the component rates.


The independent increment property renders Poisson a memoryless process. This, in turn, greatly simplifies queueing problems involving Poisson arrivals. In spite of the analytic simplification, Poisson processes do have a significant modeling






28


drawback the incapacity of capturing the autocorrelation function of {I}, which to a large extent explains the phenomenon of traffic burstiness.

3.1.2 Markov Process Models

Unlike Poisson process models, Markov process models introduce dependence into the random sequence {I,} so that they can potentially capture traffic burstiness. Consider a continuous-time Markov process X(t) with a discrete state space i E {0,1, 2,..., m}. Consequently, a set of random variables X : {X4} form a Markov chain, where the dependency extends backwards only one unit in time. In other words, the way in which the entire past history affects the future of the process is completely summarized in the current state of the process. This property is so called "Markov" or "memoryless" property.

Figure 3.1 illustrates an example of 2-state Markov chain where X stays in a state 0 for an exponentially distributed holding time (required to satisfy the Markov property) with parameter A0, and then jumps to state 1 with probability a. In steady state, we have the state equation






= Q.[ P01 (3.3) where po and pl are the steady-state probability of X0 and X1, respectively, and Q is called a transition probability matrix. Note that in a simple Markov traffic model, each jump of the Markov process X is interpreted as signaling an arrival, so interarrival times are exponentially distributed with their rate parameters dependent on the state from which the jump occurred. This situation results in dependence among interarrival times as a consequence of the Markov property.






29







0 1







Figure 3.1: 2-state Markov chain

3.1.3 Markov-Modulated Process Models

Markov-modulated process models constitute an extremely important class of traffic models. The idea is to introduce an explicit notion of state into the description of a traffic stream an auxiliary Markov process is evolving in time and its current state controls (modulates) the probability law of the traffic mechanism..

Now assume that while X is in state i, the probability distribution function (pdf) of arrivals is completely determined by i, and this holds for everyi E {0, 1, 2,..., m}. Note that when X undergoes a transition to state j E {0, 1, 2, ..., m}, then a pdf for arrivals takes effect for the duration of state j, and so on. Thus the pdf for arrivals is modulated by the state of X.

The most commonly used Markov-modulated process is Markov-Modulated Poisson Process (MMPP) which combines the simplicity of the modulating (Markov) process with that of the modulated (Poisson) process. In this case, the modulation mechanism simply stipulates that in state i of X, arrivals occur according to a Poisson process at rate Ai. As the state changes, so does the rate. As a simple example, consider a two-state MMPP model, where one state is an "on" state with an associated positive Poisson rate, and the other is an "off" state with associated rate zero (such models are also known as interrupted Poisson). These models have been widely






30


used to model voice traffic sources, where the "on" state corresponds to talk spurt, and the "off" state corresponds to silence.

3.2 Self-similar Processes

One of the most remarkable aspects about self-similarity in '90s is probably that it has been consistently observed in recent studies and measurements in network traffic [27, 5, 29]. Self-similarity is manifested in such a way that there is no natural length for a "burst" across all time scales. In other words, a bursts consists of bursty subperiods separated by less bursty subperiods. For instance, fractal images in computer graphics and snowflakes are probably two of the most well-known objects exhibiting self-similar or "fractal-like" phenomena where each small portion of them can be viewed as a reduced-scale duplicate of the whole. Some aspects of "selfsimilarity," coined by Mandelbrot [30], also appears in hydrology, economics, and communications and is related to so called "1/f noises." [31]

Self-similar traffic in networks behaves so differently from voice traffic that none of the traditional Poisson-based traffic models described in Session 3.1 is able to capture the fractal behaviors. It is also shown that the generally accepted argument that aggregate traffic becomes smoother as the number of traffic sources increase has very little to do with the reality in Ethernet LANs [27]. On the contrary, the burstiness of LAN traffic typically intensifies as the amount of active traffic increases.

Figure 3.2 illustrates the empirical trace collected from a WWW server at a typical Internet service provider (ISP) in June 1997. The breakdowns of packet arrival rate are plotted according to time scales ranging from 0.06 second to 60 seconds. One can observe the similar appearance of "burstiness" across a wide range of time scales, where the plot on the top corresponds to the complete trace with a resolution of 60 seconds, the second from the top an extraction of the first one (1/10th) but with ten times finer resolution (6 seconds), and so on. It is evident that the burstiness persists over different time-scales, and that the traffic is statistically self-similar.







31





















1400



CO to


1 time (60 sec) .....600 250



...........

1time (6 sec) 600
40






16 1 tim e (0.6 sec) 6..0.... .......... 60







1 time (0.06 sec) 600




Figure 3.2: Self-similarity of WWW traffic






32


3.2.1 Definitions

A wide-sense stationary (WSS) random process X(t) is defined as exactly selfsimilar [1] if a small portion of X(t) can exactly reproduce a large portion of X(t) by scaling properly. What is of more interest to network performance analysis, however, is the so called statistical self-similar [1]. A WSS random process X(t) is (statistically) self-similar if each smaller portion shares statistical properties of larger portion of X(t) after properly scaled. A more explicit definition of self-similar in [32] is defined below.


Definition 1 X(t) is self-similar with "Hurst parameter" H E (0, 1], if




X(t)la -HX(at), V a (3.4) and

X(t + 7) X(t)X(7-) X(O),V r (3.5) where denotes the equivalence of distribution.

By this definition, a self-similar process has stationary increments as indicated in (3.5). In addition, this process is also distribution invariant under appropriate scaling in time and space in a sense that the self-similarity can be observed by comparing small bursts in a small region with larger ones, with smaller ones superimposed, over a larger region of the horizontal axis. This distribution invariant characteristic is observed in an empirical trace show in Figure 3.2 as well. The variations of the definition of self-similarity which lead to more important properties will be described as follows.

Let X(t) be a WSS random process with mean p, variance a2 and autocorrelation function r(r). In particular, we assume that r(r) is of the form


r(7) = r2-2HS(7), as T oo00 (3.6) = r S(r), as 7 -- 00 (3.7)






33


where 0 < H < 1 and S(r) is slowly varying at infinity, namely lim.oo S(rx)/S(T) = 1, for all x > 0. Denote X(m) the new process obtained by averaging the original series X in non-overlapping sub blocks of size m. That is:

1 m-1
X(m) (k) = Xk-i (3.8) m i=0

Note that for each m, X(m) defines a WSS random process. Definition 2 Denote r(m)(T) the autocorrelation function of X(m). The process X(t) is "exactly second-order self-similar" [1] with Hurst parameter H if r(m)(7) = r(7), for all m = 1, 2, ... (3.9) In other words, X is exactly second-order self-similar if the aggregated processes are indistinguishable from X with respect to their first and second order properties. With the relaxation of (3.9), we arrive at the following definition. Definition 3 X is "asymptotically self-similar" with Hurst parameter H if

r(m)(r) = 1/2. [(7 + 1)2H + ( 1)2H 272H], as m -- 00. (3.10)

Specifically, for large m, the aggregated process X(m) has a fixed autocorrelation structure determined only by H.

The most striking feature of self-similarity is that the correlation functions of the aggregated processes do not decay to 0 as m --+ oo. This feature is in contrast to traditional models, all of which have the property that the correlation function of their aggregated processes decline as m -+ oo. i.e., r(m)() = 0, as m for T= 1,2,3,.. (3.11) Two of the most well-known self-similar processes are fractional Gaussian noise (fGn) and its incremental version, fractional Brownian motion (fBm), which will be discussed in detail later in this chapter.






34


3.2.2 Self-similarity and Long-Range Dependence

A process satisfying (3.6) is said to exhibit long-range dependence (LRD). Such processes are characterized by an autocorrelation function r(T) which decays hyperbolically as 7 increases. This implication shows that r(r) is non-summable, i.e. E' r(r) = oo which contrasts with many conventional short-range dependent (SRD) processes such as Poisson-based processes which have an autocorrelation function r(T) which decreases exponentially as 7 increases and is consequently summable, namely, E' r(7) < oo00. Furthermore, the non-summability of the correlations captures the intuition behind LRD, namely, that while individual values of r(r) are small for large 7, their cumulative effect is of importance, and produces features which are drastically different from SRD processes. Proposition 1 An asymptotically self-similar process X(t) with Hurst parameter H exhibits LRD as 0.5 < H < 1, and SRD as 0 < H < 0.5. Proof:

From (3.10), the autocorrelation function of X(m) given Hurst parameter H is

r(7) = 1[(7 + 1)2H + (7 1)2H 27-2H]. We have




r(r) 1 2H-2 2 1 2H +2( 2H -27
2 7 7
= 2T2H-272[(1 + T-1)2H + (1 -1)2H 2]}, (3.12)




(1 + T- )2H = 1 + 2Hr-1 + H(2H 1)T-2 +-.., (3.13) and

(1 T-l)2H = 1 2HT-1 + H(2H 1)r-2 +.... (3.14)

Combining (3.12), (3.13) and (3.14), we obtain







35





0.8


0.6

H=0.9
0.4
H=0.8

0.2 H=0.7 ......




H=0.1
-0.2
H H= 0.1

-0.4


-0.6
10 101 102 103 104





Figure 3.3: Autocorrelation function and long-range dependence




r(r) = 1 2H-2[2H(2H 1) + f(7)], (3.15) where f(-) = a1r-2 + a27-4 + Since f(-) = 0 as 7 00, we have



r(r, H) = H(2H 1)T2H-2. (3.16) From (3.16), it is obvious that r(T) is summable if H < 0.5, and non-summable, otherwise. E

Figure 3.3 depicts r(T) for several values of H. Of particular notice is r(T) decays to 0 as r -+ 00 much slower when H > 0.5 than when H < 0.5. As a result, r(r) tends to 0 so slowly that E' r(T) = oo, which leads to the LRD of X(t).






36


3.2.3 Self-similarity and Hurst Effect

Self-similar processes also provide an elegant explanation of an empirical law known as Hurst's law or the Hurst effect. For a given set of observations X1, X2, ..., X, with sample mean X(n) and sample variance S2(n), the "rescaled adjusted range" or "R/S statistic" is given by

R(n)/S(n) = max(0, di, d2,..., dn) min(0, di, d2, ..., dn) (317) S(n)

where


dk=X +X2+...+Xk-k X(n), k= 1,2,...,n. (3.18) Hurst [33] found that many naturally occurring time series well represented by the relation

E[R(n)/S(n)] a nH, as n oo (3.19) with Hurst parameter H normally around 0.73, and a a finite positive constant independent of n. However, if the observations come from a short-range dependent (SRD) process, then it has been shown in [31] that E[R(n)/S(n)] b n0.5, as n -- oo (3.20) with b another finite positive constant independent of n. A Matlab script for Hurst parameter estimation based on R/S statistic is included in Appendix A.1.

3.2.4 Self-similarity and Slowly Decaying Variances

Another important feature of self-similar processes is that the variance of the arithmetic mean, ft, decreases slower than the reciprocal of the sample size m. Research [34] shows that (3.6) is equivalent to var(X(m)) -* am-0, as m 00, (3.21)






37


where a is a finite positive constant independent of m, and 3 = 2 2H. On the other hand, for short-range dependent processes for which r(m) (k) = O,as m -- 00, it can be shown as well that

var(X(m)) -4 bm', as m -- 00, (3.22)


where b is a finite positive constant independent of m. The Hurst parameter estimation based on (3.21) is called Variance-time plot (vtp). A Matlab script for vtp is included in Appendix A.2.

3.3 Modeling of Self-Similarity

For decades, self-similarity has been a phenomenon that has been studied extensively in several other fields including hydrology [31] and economics [35]. Because of this exhaustive research, several formal mathematical models have already been developed which exhibit self-similarity. However, due to the abstract nature of these models, a practical interpretation for the self-similarity is difficult to find. Two such models are presented in this section: the fractional Gaussian noise (fGn), and the fractional autoregressive integrated moving average models (f-ARIMA). We also presented fraction Brownian motion (fBm), the companion of fGn, and a construction based on aggregating many simple renewal rewards processes. The latter is of particular significance to this research as it provides a practical interpretation for the self-similarity found in Ethernet LAN traffic.

3.3.1 Fractional Brownian Motion

The ordinary Brownian motion, B(t), describes the movement of a particle in a liquid, subjected to collision and other forces. This motion is a self-similar process with Hurst parameter 1/2 and independent Gaussian increments such that,


B(t)-ltl/2B(1) (3.23)






38

and
E[(B(tl + 7) B(tl))B(tl)] = 0. (3.24) From (3.23) and (3.24), it follows that

E{[B(tl + 7) B(tl)]2} = jr| E[B2(1)] (3.25) Mandelbrot [31] defines fractional Brownian motion (fBm) with Hurst parameter H as being the moving average of dB(t) in which past increments of B(t) are weighted by the kernel (t s)H-1/2,



BH (t) = F(H 1/2) [(t- 2 (s)Hl-1/2]dB() + s)H-1/2dB(s), (3.26)
where F denotes the gamma function: F(r + 1) = rF(r) and F(k + 1) = k! as k being an integer. By definition in (3.4), we have BH(t) -tHBH(1). (3.27) Several important properties of fBm are as follows.

1. BH(0) = 0

2. E{BH(t)} = 0

3. E{(BH(tl + 7) BH(tl)) = 2H E{B(1)}

4. E{)B(tl)BH(t2)} = (It12H + It2 12 -It t212H) E{B2(1)}


3.3.2 Fractional Gaussian Noise

The increments of fBm, G(n), form a stationary sequence called fractional Gaussian noise (fGn).






39



G(n) = BH(n + 1) BH(n) (3.28) A fractional Gaussian noise (fGn) is a stationary Gaussian process with mean pI, variance a2, and autocorrelation function r(r) = (1/2)(17 + 112H 2|r2H + 17 112H) (3.29) From Proposition 1, we prove that r(r) -+ H(2H 1)1,2H-2 as r -* oo, which means fGn exhibits LRD It can also be shown that the aggregated processes X(m) all have the same distribution as X for all 0 < H < 1. Thus, by (3.6), fGn is exactly self-similar with Hurst parameter H and possesses LRD as long as 1/2 < H < 1.

Fractional Gaussian noise was originally introduced in 1968 by Mandelbrot and Van Ness [31]. The applicability of fGn in traffic models is somewhat limited by its strict autocorrelation structure. This limitation makes it less suitable for modeling traffic streams which have very strong short-range dependence; for some phenomenon, however, fGn can be a reasonable approximation. It should also be noted that although fGn is a Gaussian process, an fGn sample path can be transformed into a sample path with arbitrary marginal distribution while maintaining the Hurst parameter H [36].

3.3.3 Fractional ARIMA(p,d,q) Processes

Autoregressive-moving average (ARMA) processes are often used for modeling empirical time series. Let p and q be non-negative integers, then the sequence {X,} is called ARMA(p,q) if it satisfies the equations p q
X, =E akXn-k + bkwn-k, (3.30) k=1 k=O
where {Wk : k = 0, 1, 2,...} are i.i.d random variables and are assumed to be either Gaussian or non-Gaussian. The {ak} are referred to as "AR parameters" while {bk are called "MA parameters." For convenience, denote






40



#(D) = 1 ajD -...- apDP (3.31) and

O(D) = bo biD ... bqD', (3.32) where D is a delay operator, i.e., DkX, = Xn-k. It follows that (3.30) can be rewritten as


Q(D)X, = E(D)ws, (3.33) or

X, = -1(D)(D)wn. (3.34) We now turn to fractional ARIMA time series. Let A be the difference operator, defined by AX, = X,- X,_1 = (1 D)X,. A fractional ARIMA(p, d, q) process is defined as a stochastic process X = (Xk : k = 0, 1, 2, ...) represented by 4(D)AdX = O(D)wk. (3.35) Note that the parameter d in Ad is allowed to take fractional values, either positive or negative. Logically the fractional difference operator follows as

00
Ad = (1- D)d= E C(d, k)(-D)k, (3.36) k=1
where


k k-j-1 F(k-d)
j=1 j (-d)F(k + 1) (3.37) For example, if p=q=0 and d is a non-negative integer, then AdX = wn describes a model where X,, differenced d times, yields a sequence of i.i.d. random variables wn. As a consequence, ARIMA(0,1,0) characterizes a random sequence generated by random walk. Of particular notice is it has been shown in [34] that for d E (0, 1/2)






41


a fractional ARIMA(p, d, q) process satisfies equations Eq. 3.6, which makes it an asymptotical self-similar processes with LRD and Hurst parameter H = d + 1/2.

Fractional ARIMA(p, d, q) processes were first introduced in [37]. One advantage of fractional ARIMA(p, d, q) processes over fractional Gaussian noise is their flexibility with respect to modeling both short-range and long-range dependence, making them better suited to modeling phenomenon such as VBR video traffic [36].

3.3.4 Self-Similarity Through Aggregation

Consider a number of independent sources which alternate between the states "on" and "off" (renewal-rewards processes) where the amount of time spent in each state is randomly distributed with a heavy-tailed distribution. An example of such a heavy-tailed distribution is the stable Pareto distribution (see Appendix B of [5]) with parameter 1 < a < 2. If we construct a new sequence consisting of the number of sources which are "on" at points in time separated by a fixed length interval then it has been shown in [1] that the resulting sequence will be asymptotically self-similar.

The notion of generating a self-similar process by aggregating a number of simple renewal rewards processes provides intuition into the self-similarity of Ethernet traffic as well as into the nature of a single source on an Ethernet network. If we consider each workstation (or more likely each user) attached to an Ethernet network as a renewal-rewards process where the process is "on" if the workstation is communicating over the network and "off" otherwise, then we see that the amount of traffic on the network can be thought of as the aggregation of a number of renewal rewards processes. This also leads to the conclusion that the communication bursts produced by individual users have a heavy-tailed distribution. This conclusion is supported by in depth analysis of Ethernet [1] and wide area network traffic traces [38].






42


3.4 Self-similar Traffic Trace Generators

While most network management tools such as HP's OpenView and IBM's NetView have built-in capabilities to monitor and gather trace data, not everyone has access to such tools. One of the most popular monitoring tools designed as user commands is "tcpdump," which is useful to get an overview of what type of or how many packets are on a network for a given time period. The problem is that not many people feel comfortable reading and recording them, mostly due to the concerns of individual privacy or network security. However, to do simulation studies, a lot of trace data is needed, and gathering and storing this data not only is time consuming but also requires huge storage space. What is needed is a fast method of generating synthetic traffic traces for use in simulations. Currently, a number of approaches to generate self-similar sample paths are available, and we introduce several popular ones in the following sections.

3.4.1 Random Midpoint Displacement Algorithm

The most well-known algorithm for generating fBm is the "Random Midpoint Displacement" (RMD) algorithm. RMD works by progressively subdividing an interval over which to generate the sample path. At each division, a random displacement drawn from a Gaussian distribution is used to determine the value of the sample path at the midpoint of the subinterval. Self-similarity comes about by appropriately scaling the variance of the displacement. A Matlab script for RMD is included in Appendix B.1. Examples of the resultant fBm and fGn using this script are depicted in Figure 3.4, where fBm and fGn are shown in the first and the second column, respectively. In this figure, various Hurst parameters with a value of 0.1, 0.5, and 0.9 are shown in the first row, the second row and the third row, respectively. Note that fGn is simply the increments of fBm. Also note that negative dependence and LRD are observed as H=0.1 and H=0.9, respectively.






43










fBm fGn
10 5


H=0.1 5 0
0

-51 -5
0 500 1000 0 500 1000


2 0.5


H=0.5 0
-2

-4 -0.5
0 500 1000 0 500 1000


4 0.1

2
H=0.9 0
0

-2 -0.1
0 500 1000 0 500 1000





Figure 3.4: Fractional Brownian motion and fractional Gaussian noise






44


3.4.2 Aggregation of Renewal Processes

A self-similar trace generator described in [39] consists of aggregating a number of renewal rewards processes to obtain a self-similar process. This approach has the problem that one must trade off speed for the degree of self-similarity, as increasing the Hurst parameter requires that the number of renewal rewards processes be increased. Conversely, one advantage of this method is that it can be efficiently implemented on a parallel computer. Leland and et al [39] report taking 3-5 minutes to generate a trace of 100,000 points on a massively parallel computer with 16,384 processors using this method.

3.4.3 M/G/oo with Heavy-tailed Distributed Service Time

Consider an M/G/oo queue, where customers arrive according to a Poisson process and have service times drawn from a heavy-tailed distribution. In this model, the value of the sample path at time t is the number of customers in the system at time t. This model produces sample paths which are asymptotically self-similar. The drawback of this approach is that, once again, one must trade off speed for the degree of self-similarity, but unlike the previous approach, this method is not easily parallelized.

3.4.4 ARIMA

This method described in [2] generates sample paths from a fractional ARIMA process which are asymptotically self-similar. This algorithm has the major drawback that it requires O(n2) running time to generate a sample path of length n. The authors of [2] report taking 10 CPU hours to generate 171,000 points using this method.






45


3.4.5 Approximation of Power Spectrum

The method described in [38] is based on the Fast Fourier Transform. The mathematics of this method are beyond the scope of this research, but the strategy behind the method is to generate a sequence of complex numbers corresponding to the power-spectrum of fractional Gaussian noise. The inverse Discrete Fourier Transform is then used to obtain the time-domain counterpart of this power-spectrum. Because autocorrelation and power-spectrum form a Fourier pair, the resulting process is guaranteed to have the autocorrelation structure (and, hence, self-similarity) of fractional Gaussian noise. This method has the property that it is fast (generating 262,144 points takes 80 seconds on a SPARCstation IPX) and that it does not suffer from the biases associated with the Random Midpoint Displacement algorithm. Because it is based on the FFT, this method is referred to by its authors as the FFT method of generating self-similar sample paths. A Matlab script based on this method is included in Appendix B.2.















CHAPTER 4
CONGESTION AND FLOW CONTROL

The B-ISDN, which is based on the ATM technique, is designed to transport a wide variety of traffic classes satisfying a range of transfer capacity needs and network performance objectives. The underachievement, if any, is usually caused by network congestion which could happen whenever the overall instantaneous input rate is greater than the link capacity for a certain period of time, i.e.



A, > C (4.1) where A1 represents the instantaneous input rate of an active source, and C the link capacity. To be more general, congestion in ATM networks can be caused by statistical fluctuation of traffic flows or network failures [20]. In light of the diverse traffic characteristics in ATM networks, the design of a scheme to adequately control the data flow into ATM networks so as to avoid congestion becomes a critical factor for the success of B-ISDN. Aside from the primary role of congestion and flow control schemes to protect the network and the user in order to achieve network performance objectives, an additional role is to optimize the use of the network resources so as to achieve better utilization efficiency.

4.1 General Framework for Congestion Management and Control

For a congestion management and control scheme to be able to support a set of QoS classes sufficient for all foreseeable services and maintain minimum network and end-system complexity while maximize network utilization, the following functions





46






47


form a framework for managing and controlling congestion in ATM networks and may be used in appropriate combinations [20].

Resource Management (RM) [14, 40, 41]: RM provides for allocating network

resources in order to separate traffic flow according to the QoS on traffic contract. Two critical resources in ATM networks are buffer space and trunk

bandwidth.

Connection (or Call) Admission Control (CAC) [42, 43, 44]: CAC defines a

set of action taken by the network during call set-up or re-negotiating phase in order to determine whether a VCC or VPC request can be granted or not.

Apparently routing is part of CAC actions. One important class of applications of ATM networks is interactive programs such as videoconferencing. In order to provide QoS requirements for all connections, CAC is an essential mechanism.

Traffic Shaping (TS) [45, 46]: A key element of the traffic contract from the user

perspective is that the sequence of cells can be sent to the network after a desired modification of the traffic characteristics and is still compliant with the traffic parameters in the contract. In other words, the user equipment can process the source cell stream such that the resultant output to the network is conforming to the traffic contract. The most popular traffic shaping algorithms as proposed in literatures include Peak Cell Rate Reduction (PCRR) and Buffering. PCRR can be achieved by operating the sending terminal at a peak rate less than that in the traffic contract, reducing the possibility of conformance violation.

Buffering operates in combination with the leaky bucket algorithm to ensure that cells will not violate the traffic parameters of the contract by buffering

cells until leaky bucket would admit them.

Priority Control [47]: By using Cell Loss Priority bit, the user may generate

traffic flows with different priorities. An ATM switch may selectively discard






48


cells with low priority if necessary to protect the QoS of cells with high priority

as much as possible.

Usage Parameter Control (UPC) [48] is defined as the set of actions taken by

the network to monitor and control traffic in terms of traffic offered and fairness of ATM connections. The main purpose is to protect network resource from malicious or unintentional misbehaviors, which can affect the QoS of other already established connections, by detecting violations of negotiated parameters

and taking actions accordingly.

Feedback Control [6, 13, 49, 8] is defined as the set of actions taken by the

network and by the users to regulate the traffic submitted on ATM connections according to the state of switch such as the dynamic value of Cell Loss Rate,

Cell Transfer Delay or buffer occupancy.


One way to classify congestion control schemes is by the layer of ISO/OSI reference model at which the scheme operates [50]. For example, there are data link, routing/networking, and transport layer congestion control schemes. Typically, a combination of such schemes is used. For sporadic congestion, CAC is used to route according to load level of the links and to reject new connections if all paths are highly loaded. For congestion lasting less than the duration of connection, an end-to-end control scheme can be used. For example, during connection setup, the sustained and peak rate may be negotiated. A leaky bucket algorithm may be used later by the source or the network to ensure that the input meets the negotiated parameters. Such "traffic shaping algorithms" are open loop in the sense that the parameter cannot be changed dynamically if congestion is detected after congestion. On the contrary, in a closed loop scheme, sources are informed dynamically about the congestion state or resource usage of the network and are asked to adjusted their input rate accordingly. In general, this information is carried by acknowledgment or RM cells via hop-by-hop






49


or end-to-end feedback. Hop-by-hop feedback requires each intermediate switching node to send the information to its previous stage and is, therefore, more suited for services which feature short-term overloads. On the other hand, end-to-end feedback is more effective for services with long-term overloads because of its capability to react to the feedback in a more promptly manner.

4.2 Flow Control Schemes with Feedback

Most congestion control schemes for ATM networks consist of adjusting the input rates to match the available link rate in attempt to improve the network utilization efficiency via feedback. Flow control schemes using feedback are spearheaded by two approaches: rate-based, and credit-based flow control scheme. After a series of intensive debates in ATM forum, the former was finally selected as the standard of the flow controller in B-ISDN mainly thanks to its higher cost effectiveness compared to the latter.

4.2.1 Credit-Based Scheme

This is one of the two leading approaches and also the first one to be proposed, analyzed, and implemented. The approach consists of per-link, per-VC (Virtual Channel), and window flow control. Each link consists of a sender node and a receiver node. Each node maintains a separate queue for each VC. The receiver monitors queue length of each VC and determines the number of cells that the sender can transmit on that VC. This number is called "credit". The sender transmits only as many cells as allowed by the credit. In the case of a single active VC, the credit must be large enough to allow the whole link to be full at all times. In other words:



Credit > Link Cell Rate x Link Round Trip Delay, (4.2) where one can compute the link cell rate by dividing the link bandwidth in Mbps by the cell size in bits.






50


The credit-based scheme as described so far is also referred to as "Flow Controlled Virtual Circuit (FCVC)" scheme. This initial static version has two problems. First, if the credits are lost, the sender will not know it. Second, each VC needs to reserve the entire round trip worth of buffers even though the link is shared by many VCs. These problems were solved by introducing a credit resynchronization algorithm and an adaptive version of the scheme.

The credit resynchronization algorithm consists of both sender and receiver maintaining counts of cells sent and received for each VC and periodically exchanging these counts. The difference between the cells sent by the sender and those received by the receiver represents the number of cells lost on the link. The receiver reissues that many additional credits for VC.

The adaptive FCVC algorithm consists of giving each VC a fraction of the round trip delay worth of buffer allocation. The fraction depends on the rate at which the VC uses the credit. For highly active VC, the fraction is larger, while for less active, the fraction is smaller. Inactive VCs get a small fixed credit. If a VC does not use its credits, its observed usage rate over a period is low and the VC gets smaller buffer allocation (and, hence, credits) in the next period. The adaptive FCVC reduces the buffer requirements considerably, but also the full capacity of the link, even if there are no other users.

4.2.2 Rate-Based Scheme

The original proposal for a rate-based approach consists of end-to-end control using a single-bit to control the network. In the proposal, the switches monitor their queue lengths and, if congested, set the explicit forward congestion indication (EFCI) in the cell headers. The destination monitors these indications for a periodic interval and sends an RM cell back to the source. The sources use an additive increase and multiplicative decrease algorithm to adjust their rates. This particular algorithm






51


uses a "negative feedback" in the sense that RM cells are sent only to reduce the rate but no RM cells are required to increase the rate. The problem is that RM cells may be lost due to heavy congestion in feedback, and, consequently, the sources will keep increasing their sending rate and eventually overload it. This drawback is enhanced by using a positive feedback where it would require a RM cell to increase the rate but not on decrease. In a more enhanced version, RM cells are sent for both increase and decrease.

An example of rate-based flow control scheme using a negative feedback is "Backward Explicit Congestion Notification" (BECN) scheme. This method, presented by N.E.T [51, 52], consists of switches monitoring their queue lengths and sending an BECN cell back to source if congested. The sources reduce their rates by half on the receipt of the BECN cell until a pre-defined minimum rate is reached. If no BECN cells are received within a recovery period, the rate for that VC is doubled once each period until it reaches the peak rate. Note that when the minimum rate is reached, any further receipt of BECN cell will force the source to stop transmitting. To achieve fairness, the source recovery period was made proportional to the current level of transmission rate so that the lower transmission rate, the shorter the source recovery period [53].

When supporting TCP traffic over ATM networks, a slow-start mechanism may be used jointly with the BECN scheme to avoid packet losses immediately after the initial transmission in a overloaded situation. Specifically, instead of starting the transmission at its peak rate, the source starts with the minimum rate and then recovers to its peak rate gradually by applying the same recovery procedure as in the BECN scheme.






52

4.3 Impropriety of Poisson-Based Traffic Models

This session presents the simulation efforts to justify the performance of the flow control schemes introduced in the previous section by using Poisson process traffic models and self-similar process (SSP) traffic models. The RMD algorithm is used to generate a self-similar sample path. Since fGn has zero mean, it can not be directly used to represent a TCP packet size. However, since a linear transformation does not affect Hurst parameter, it may be used to make the sample paths eligible for a packet size. Note that due to the re-transmission of erroneous packets in TCP, any single cell loss of a packet in ATM networks will result in the discard of any other successfully transmitted cells of this packet. The effective throughput taking into account the cell discard is so-called goodput.

4.3.1 Network Configuration

Figure 4.1 shows the network configuration of the simulation scenario where two CPS-100 ATM switches are connected to each other by a trunk link at a speed of DS-3 (44.5 Mbps). Eight CPEs are attached to each of the two switches. We assume a one-to-one correspondence among each pair of transmitting and receiving CPEs. In other words, a virtual path or channel connection is established between each CPE attached to CPS 100-A and the corresponding CPE attached to CPS 100-B. The RMD algorithm is used to generate synthesized self-similar traffic with a target Hurst parameter at each CPE attached to CPS 100-A. The self-similar data flow is then destined for the CPS 100-B through the trunk link. Note that the CPEs attached to CPS100-B send only RM cells with congestion information through feedback. As active traffic sources aggregate on the outgoing trunk port of CPS100-A, the egress buffer will build up eventually if there are too many arriving cells in a short period. The egress buffer is, therefore, a performance bottleneck. Due to balanced traffic and symmetric network topology, the same value of effective throughput, namely,






53





TX-2 Rcv-2 TX-3 Rcv-3


TX-4 acv-4







TX-7 Rcv-7 TX-8 Rcv-8


Figure 4.1: Network scenario for Poisson vs. Self-similar traffic models

goodput, is expected from each of the eight connections. The maximum goodput that may be achieved in this scenario is 0.125.

The Hurst parameter of the generated sample paths is estimated by Variancetime plot (vtp) as shown in Figure 4.2. Note that the RMD algorithm renders a Hurst parameter which is slightly smaller than the target value 0.9. In addition, we noticed that the sample path generated by Poisson process yields a Hurst parameter 0.5 which is quite different from the range of [0.70 0.90] empirically found on real networks [27]. In our simulations, for both Poisson process and SSP traffic sources, the egress buffer size is equal to 500 cells and offered traffic load is maintained at the same level to facilitate meaningful comparisons.

4.3.2 Effective Throughput Comparisons

Figure 4.3 shows the goodput of VC-1, i.e., the virtual connection between Tx1 and Rcv-1, for various flow-control schemes where Poisson process traffic models are used. Despite the obviously superior performance of the credit-based scheme as






54






-0.5 0000000H=0. O




S-1.5

EO
S-2 o:SSP sample
+:Poisson sample +
0 +
-2.5 +

-3
0 0.5 1 1.5 2 2.5 3 logl,(Aggregated level (m))


Figure 4.2: Variance-time plot for synthesized sample paths

verified by the results shown in this Figure [11], the ATM Forum has selected ratebased flow control scheme to be used for flow and congestion control in ATM networks. Thus, in studying the effects of self-similar traffic on the network performance, we concentrate on the BECN scheme with or without the slow start option. Note that in this session, "rate-based scheme I" and "rate-based scheme II" are referred to the BECN scheme without and with slow-start, respectively.

Figure 4.4 and Figure 4.5 show the goodput of VC-1 for the BECN scheme with and without slow-start, respectively. In both figures, Poisson process as well as SSP with H=0.7 and 0.9, are used as the traffic models. The values of goodput at steady state are listed in Table 4.1 where the first column is the traffic model for the sources, the second and fourth columns are the flow control schemes, and the third and fifth columns are the percentage of over-estimation of goodput by using Poisson process instead of SSP traffic models. Evidence shows that using a Poisson process






55


Traffic model rate-based I A% rate-based II A%
Poisson 0.0934 0.0973
SSP, H=0.7 0.0869 7.2% 0.0914 6.5%
SSP, H=0.9 0.0825 13.2% 0.0868 12.1%

Table 4.1: Performance comparison of Poisson process and SSP traffic models


0.14
credit-based


0.1 -/[ rate-based II

0.08
0.0 rate-based I

0
o 0.06

0.04 0.02


0 2 4 6 8 10 simulation time (sec)

Figure 4.3: Performance comparison of different flow control schemes


model for Self-similar traffic with H=0.7 renders 7.2 % and 6.5% of over-estimations of the goodput for the BECN scheme without and with slow-start, respectively. While using a Poisson process model for Self-similar traffic with H=0.9 results in 13.2 % and 12.1% of over-estimations of the goodput for the BECN scheme without and with slow-start, respectively. These results demonstrate the failure of Poisson processes as traffic models for traffic exhibiting self-similarity.


4.4 Classical Control Theory and Congestion Control

For a system as shown in Figure 4.6, classical control theory is usually valuable as a source of insight in the design of feedback controllers. In a feedback flow control






56




0.1


0.08
Poisson process model .- 0.06 SSP model, H=0.7 O SSP model, H=0.9
0
0
S0.04



0.02


0
0 2 4 6 8 10 simulation time (sec)



Figure 4.4: Performance comparison of rate-based scheme I




0.1


0.08
Poisson process model

S 0 .0 6 .. .............................................................................. ..... S S P m odel, H =0 .7

O SSP model, H=0.9
0.04



0.02


0
0 2 4 6 8 10 simulation time (sec)



Figure 4.5: Performance comparison of rate-based scheme II






57



Uncertainty

b' Controller Network b



Feedback



Figure 4.6: Generic Feedback Control System system, this figure is the type of system used to ensure that the network, the controlled object characterized by input u, internal states x, output y, and unavoidable uncertainty should perform in a desired manner.

Several important insights from classical control theory are as follows.


1. A feedback controller must combine dual function of estimation and of control.

The estimation reduces uncertainty about current state whose current values

are needed in order to determine current control action.

2. Feedback is the means by which control is achieved in the presence of uncertainty. Therefore, any feedback system should benefit from a design approach

which take explicit account of uncertainty.

3. Two different types of uncertainty can affect controlled objects: uncertainty

about current values of the internal states; and uncertainty about future values resulting from uncertainty about controlled object dynamics governing the

evolution of future state.


The concept drawn from the classical control theory will serve as a basis for the proposed flow control schemes in Chapter 6.99















CHAPTER 5
PREVENTIVE FLOW CONTROL SCHEME FOR SELF-SIMILAR TRAFFIC Recent measurement studies have shown that the burstiness of the traffic in LANs or WANs is associated with long-range dependence (LRD). In Chapter 3, we showed that LRD can be efficiently characterized by a queueing model with selfsimilar arrival processes such as fractional Gaussian noise (fGn), or autoregressive integrated moving average (ARIMA) processes. As indicated in [1], the fGn provides the simplest way to model the LRD and self-similarity. In this chapter, we will, consequently, consider fGn as the arrival process in an ATM network environment.

To avoid the congestion and buffer overflow resulted from bursty traffic, we propose a preventive flow control scheme which consists of a new call admission control (CAC) algorithm. This CAC algorithm uses the Hurst parameter as an index of burstiness and to characterize the self-similar nature of aggregate traffic in virtual paths of ATM networks. To facilitate the efficiency of CAC for self-similar traffic, we derive the cell loss probability for a queueing system with finite buffer and fGn arrival process. Note that the derivation is served as an extended version of the work proposed by I. Norros [54]. The numerical results reveal that the proposed CAC algorithm outperforms the peak rate allocation scheme by rendering higher statistical multiplexing gains (SMG). In addition, the algorithm provides less deviation from the target cell loss probability than other statistical allocation scheme such as "Equivalent Capacity".








58






59


5.1 Introduction

The Self-similar process is a realistic model for characterizing the statistical behaviors of the traffic corresponding to LANs, WANs, and VBR coded video. Self-similar processes are characterized by an hyperbolic decay of their autocorrelation function that cannot be captured by traditional Poisson-based processes. The burstiness of multimedia traffic represents the main reason behind the development of statistical multiplexing schemes for transport of heterogenous traffic over broadband networks such as ATM networks. In ATM networks, cells are switched according to information contained in their headers between ATM switches. When cells arrive at a switch, they may be stored in a buffer awaiting transmission to their new destinations. If, within a certain period of time, the number of arriving cells, which may come from different and often bursty sources, is larger than the number of cells that can be served, the buffer may overflow and cells may be lost. In video services, cell loss reduces picture quality. The design of flow and congestion control in ATM networks is, therefore, faced with the challenge of how to provide required QoS and still maintain sufficient network utilization to operate an economical viable networks.

Flow and congestion control procedures can be classified into preventive control and reactive control with the former preventing the occurrence of congestion with resource management, and the latter controlling the level of congestion based on feedback information. Both approaches have advantages and disadvantages. In ATM networks, a combination of these two approaches is used to provide effective congestion control. For instance, best effort service uses a reactive scheme while CBR and VBR are mainly based on preventive schemes.

In the previous chapter, we presented several reactive flow control schemes such as rate-based or credit-based feedback flow control schemes. In contrast, a preventive flow control generally involves the following two procedures: call admission control (CAC), and bandwidth enforcement. Since ATM is a connection-oriented, a






60


call setup procedure has to be carried out before a user starts transmitting over an ATM network. The main objective of this procedure is to establish a path between the sender and the receiver. In addition, this procedure also allocates resources in every switch along the path to accepted connections. Generally, CAC deals with the question whether or not a switch can accept a new connection. CAC may be classified into two categories, namely, statistical allocation and non-statistical allocation. Nonstatistical allocation is also referred to as so-called peak rate allocation.

The effectiveness of CAC mainly depends on appropriate traffic modeling and characterization. In this chapter, we investigate the behaviors of an ATM queueing system with a self-similar arrival process. The service process is deterministic and the arrival process is modeled by a fractional Brownian motion (fBm) process. The main goal of this chapter is to design a new effective CAC suited for services exhibiting self-similarity.

The remaining of this chapter is organized as follows. Section 2 and Section 3 describe peak rate allocation and statistical allocation scheme, respectively. In Section 4, we develop a queueing model to evaluate the cell loss probability for a system with fractional Brownian motion arrival processes. In Section 5, we propose a CAC based on the results obtained from Section 4. Numerical results are presented in Section 6, and conclusion is given in Section 7.

5.2 Peak Rate Allocation

Peak rate allocation (PRA), also called non-statistical allocation, is the most widely used CAC due to its simplicity. When a user requests a connection, the decision to accept or reject the request depends on whether its peak cell rate (PCR) is greater or less than the available rate The major drawback of PRA is that the network resources will be well under-utilized if the ratio of PCR to MCR (mean cell rate) of the connection is much larger than one. Using PRA for VBR services is one of the best examples of this situation. Given a PCR/MCR equal to R, the resource






61


utilization rate is R-1. Consequently, PRA is best suited for CBR services including PCM-encoded voice, uncompressed video and applications with low PCR such as telemetry.

5.3 Statistical Allocation

In statistical allocation, bandwidth for a new connection is not allocated based on PCR. In consequence, the sum of all peak rates may be greater than the link speed. Statistical allocation makes economic sense when dealing with bursty sources. One primary drawback of statistical allocation is its difficulty carrying out effectively, mainly due to the difficulties in characterizing an arrival process.

An example of statistical allocation is the use of "Equivalent Capacity" [55] defined as the service rate of the queue that corresponds to a cell loss probability E. By using an interrupted fluid process as the traffic model, [55] shows that given a finite buffer size K, peak rate R, MCR M, and the utilization rate p, the required bandwidth to achieve cell loss probability E is


a- K + (a K)2 + 4Kap
C = R, (5.1) 2a
where a = -In(e)p(1 p)R.

5.4 Cell Loss Probability in Self-similar Queuing Models

To evaluate the cell loss probability in an ATM connection, we studied the performance of a queueing system loaded by the composition of independent fractional Gaussian noise (fGn) processes. We described a proposed diffusion approximation for the number of arrivals in the interval (O,t], considering an heavy traffic renewal processes below.

Proposition 2 In a queueing system with renewal arrival process with mean A, the cumulative counts of arrival, a(t), can be approximated by ca(t) = It + vJiB(t), (5.2)






62


where B(t) is a standard Brownian motion, namely a Gaussian process with independent increments.

(5.2) follows from the equivalent proposition, a(t) pt
= B(t), (5.3) which can be proved by Central Limit Theorem [56].

Extending (5.2) to the case of fGn arrival processes, as suggested in [54], a(t), the number of arrivals up to time t, can be represented by a(t) = pt + vfjiLvBH(t), (5.4) where v is the variance coefficient and BH(t) is a normalized fractional Brownian motion (fBm) with zero mean and variance ItI2H. Let 6(n) denote the increment of A(t) in the interval ((n 1)T, nT]. We obtain

6(n) = a(nT) a((n 1)T)

= pfT + V1u{BH(nT) BH((n 1)T)} = 1,T + vJ-ILvG(nT), (5.5) where G(n) is a fractional Gaussian noise with zero mean and a variance equal to one. From (5.5), it then follows that E{6(n)} = jLT, (5.6) and

Var(6(n)) = vE{G2(nT)} = IvLTI2HE{G2(n)}

= ALITI2H. (5.7) From (5.5), (5.6) and (5.7), we conclude that 6(n), the increment of A(t), is also an fGn with Hurst parameter H, mean uT and variance avlTI2H. From Proposition






63


1 in Chapter 3, 6(n) possesses LRD as H > 1/2. Now let C be the fixed service rate, i.e., the link speed of an output port, then the number of arrivals up to time t, 3(t), is equal to Ct. Denoting W(t) the number of cells in the queue up to time t, then by "Reich's Formula" [57] we have


W(t) = sup[a(t) 3(t) a(x) + P(x)] x = sup[a(t) a(x) C(t x)] x sup[( C)(t x) + V/'-p(BH(t) BH(X)). (5.8) x
The following propositions are extended from the theorems originally presented in [54] to facilitate the derivation of the cell loss probability for a queueing system with fGn arrival process.


Proposition 3 W(t) is a self-similar process only if H = 1.


Proof:

W(at) = sup[(p C)(at ax) + V-I(BH(at) BH(ax))] x sup[aH((a C)al-H(t- X)+ /"jI(BH(t) BH(X))] x = H sup[a(t) a(x) [ (1- C)al-H](t x)] (5.9) x
As H = 1, we obtain W(at) = asup[a(t) a(x) C(t x)] x = a1W(t),


which implies that W(t) is a self-similar process with H = 1. E

Note that, from (5.9), when H $ 1, W(at), the number of cells in the queue up to time at has the same distribution as aH times the number of cells in a queue with the original arrival process but with service rate a (a C)a1-H






64


Proposition 4 Let p = represent the utilization rate, k the finite buffer size, then
1-p Cl-/(2Hk-l1+/H
S/2C1(2H) = constant. Proof:

Since A(t) has stationary increments, W(t) is stationary and thus has the same distribution as W(0). Substituting t with 0 in (5.8), we obtain W(0) = sup[(1- C)(-x) + V/-/(-BH(x)). (5.10) z<0

Substituting x with -t, and with the fact that BH(x) is Gaussian, we have W(0) = sup[(p C)(t) + v--iBH(t)] t>O
= max[v/-i-BH(t) (C L)t]. (5.11) t>0

Let p(k) be the cell loss probability for a queue with finite buffer size k, then we have

p(K) = Pr(W(t) > k) = Pr(W(O) > k)

= Pr(max[v/i-BH(t) (C p)t] > k) (5.12) t>O
(C /) _k
= Pr(max[BH(t) (C t] > k (5.13) t>O A 11 Since BH (t) is a self-similar process with Hurst parameter H, we have a-HBH (at)-BH(t). Substituting a with bH-l, we obtain


b-1BH(blHt) ^BH(t)

or, after changing variables, b-'BH(t)=-BH(b-1/Ht). (5.14) Let 0 denote Pr(max[BH(t) at] > b). Using (5.14), we have
t>o

= Pr(max[BH(t) at] > b) t>O






65

= Pr(max[b-BH(t) b-'at] > 1) t>O
= Pr(max[BH(b-1/Ht) b-'at] > 1) t>O
= Pr(max[BH(t) b-1+1/Hat] > 1). (5.15) t>O

Combining (5.13) and (5.15), we obtain



k (C ~i)
p(k) = Pr(max[BH(t)- ( )-1+1H t] > 1) (5.16) t>O
= Pr(max[BH(t) wt] > 1) (5.17) t>O
= f(w), (5.18)


where w = (-)-1+1/H ). Since BH(t) is Gaussian, from (5.17), it is obvious that f(w) is a monotonically decreasing function. Therefore, given a cell loss probability, say E, there exists an inverse function such that w = f-1 () =constant.

Denoting p = -, we have

w= k )-1+1/H(C L) C 1 CK-1+1/H- 1/2H I1/2H
1 PC11/(2H)k-1+1/H -12H p1/2H
= constant.




Proposition 5 Given a fractional queueing system with MCR pI,variance coefficient v, Hurst parameter H, service rate C and finite buffer size k, the approximation of the cell loss probability, p(k), is

-(C I )2H
p(k) exp( 2H k2-2H). (5.19) exP(21(HH(1- H)1-H)2 Proof:






66

From (5.13), we have

p(k) = Pr(W(t) > k)
(C ) k
= Pr(max[BH(t) __ t] _(C- ,)t + k
= Pr(max[BH(t) ] > 0) (5.20) Since BH(t) tHBH(1), substituting BH(t), we have (C-p)t + k]
p(k) = Pr(max[BH(1) (C i)t > 0) t>O tH i ] V 0 SP(B((C .)t + k
=Pr(BH()>r >n kj, t>o tH,= Q(min[(C I)t + k(5.21) t> H (5.21) where the Q-function is defined by Q(x) = t e-2/2dt. For brevity, denote (C-t+k by f(t). We have f'(t) (1 H)(C pL)t-H Hkt-H-1 f'(t) =

tH-l((1 H)(C ,)t kH)
= (5.22)


f1(t) = -H(1 H)(C )t-H-1 + H(H + 1)kt-H-2 (5.23) When t kH
Whent (C-p)(1-H) = t*, f'(t*) = 0, and

f"(t*) = (kH)-H-2[(1 H)(C li)]H+2(-H(kH) + H(H + 1)k)
f"(t*) =
(kH)-H-2[(1 H)(C pI)]H+2(Hk)
> 0. (5.24) Therefore, f(t) has a minimum when t = t*. Substituting t with t*, and using the approximation Q(x) = e-,2/2 and the theory of large deviations [58], we obtain (C- )((C k1-H )1-H + k p(k) = Q()






67


-((C )H(kH )1-H + k)2
exp[ 2] (5.25) 2puy
exp( ( H k 2-2H) (5.26) X2 (HH( H)1H)2

Given p(k) = e, the bandwidth C required to accept a connection can be computed from (5.26) and is


C = + + H(-2pv In e)l/(2H)( k )(1-1/H). (5.27) 1-H
5.5 Proposed Call Admission Control Algorithm

The simplest Connection Admission Control strategy can be obtained considering Peak Cell Rate (PCR) allocation for each traffic source. PCR can lead to no loss of cells but to a very poor link utilization. To obtain a better utilization of resources, it is necessary to accept a small degradation of the Quality of Service (QoS) while taking into account the statistical properties of actual traffics. To this end, we propose a Call Admission Control (CAC) algorithm based on (5.27). The procedure of the proposed CAC algorithm is described below.


1. Each source has to submit the traffic parameters including /I, v, and H, and

desired Cell Loss Ratio E.

2. Compute required bandwidth C according to (5.27).

3. Compare required bandwidth C with available bandwidth C'.

4. Accept the connection, if C < C', or reject, otherwise.

5.6 Numerical Results and Discussions

Assume that the offered traffic source has the parameter (u, v) = (1000, 100). To study the influence of buffer size on cell loss probability, we set the allocated bandwidth C to be 1050 cell/sec. Figure 5.1 depicts the cell loss probability vs. buffer size for several Hurst parameters, H. It is seen that when H=0.5 or H=0.6 the






68







0.8


0.6


u 0.4 ........... H=0.8
o h
0-- H=O.7

.0.2 H=0.6
0.2
o........ H=0.5


10 102 103 104 105 Buffer size


Figure 5.1: Cell loss probability vs. buffer size cell loss probability decreases significantly as the buffer size increases. However, when H=0.8 or H=0.9, increasing the buffer size merely renders a margin of improvement in the cell loss probability. This is contradictory to the results obtained from generic queueing models which do not take the Hurst parameter into account.

To study the sensitivity of the proposed CAC scheme to the changes in buffer size, we calculate the number of admitted connections for different buffer size. The parameters for the traffic source are (u, v, E) = (1000, 100, 10-s), PCR=2000 cells/sec, and the output port is a DS-3 link at the speed of 110,000 cells/sec. The number of admitted connections vs. buffer size is plotted in Figure 5.3 where the number of admitted connections decreases as the Hurst parameter increases. This is because the greater the Hurst parameter, the burstier the traffic is, and consequently the larger allocated bandwidth is required.






69


(H, e) = (0.7, 10-8) (H, ) = (0.8, 10-8) (H, ) = (0.9, 10-8) genuine H estimated CLP genuine H estimated CLP genuine H estimated CLP
0.66 3.2,10-9 0.76 1.7*10-9 0.86 4.7*10-9 0.67 5.2*10-9 0.77 1.6,10-9 0.87 3.9*10-9 0.68 4.7,10-9 0.78 3.2*10-9 0.88 6.7,10-9 0.69 6.1,10-9 0.79 4.1*10-9 0.89 6.3*10-9 0.70 7.4,10-9 0.80 6.4,10-9 0.90 8.5*10-9 0.71 7.7*10-9 0.81 6.8*10-9 0.91 7.7*10-9 0.72 8.4*10-9 0.82 8.5*10-9 0.92 9.4*10-9 0.73 8.7*10-9 0.83 9.6*10-9 0.93 8.7*10-9 0.74 1.2*10-8 0.84 1.3*10-8 0.94 9.7*10-9 0.75 1.3*10-8 0.85 1.6*10-8 0.95 1.2*10-8 Table 5.1: Robustness of the proposed CAC scheme

Assume a finite buffer size, 500 cells, and the parameter (u, v) = (1000, 100). Figure 5.4 shows the number of admitted connections given a certain cell loss probability. Note that the number of admitted connections increases significantly as the cell loss probability increases when we have a larger Hurst parameter. When H is equal to 0.5, the number of admitted connections remain nearly unchanged as the cell loss probability increases.

It is of particular interest to investigate the robustness of the proposed CAC, supposed that the submitted Hurst parameter is different from the genuine value. From Table 5.1 and Figure 5.2, we found that the requested CLP can be ensured as long as the genuine Hurst parameter is at least 0.04 less than the submitted value. This robustness is related to the fact that the CLP is derived from (5.26), which is an upper bound of the accurate value.

If the peak rate allocation scheme is used as the CAC, then the number of admitted connections is equal to the ratio of the link capacity to the PCR. Accordingly, the number of admitted connections in this case is 110,000/2000 = 55. From Figure 5.4, it is clearly that the proposed CAC algorithm achieves more admitted connections. Let us define the statistical multiplexing gain (SMG) as the ratio of the number of admitted connections by using the peak rate allocation to that by using






70



0-7
10
submitted
Hurst parameter
H=0.7
O H=0.8
H=0.9




I0. oO .

o
0


109

0.65 0.7 0.75 0.8 0.85 0.9 0.95
genuine Hurst parameter


Figure 5.2: Estimated CLP vs. genuine Hurst parameters


the proposed algorithm. The SMG reaches a maximum value 1.92 as H = 0.5. When H = 0.9, the SMG increase from 1.05 as e = 10-9 to 1.35 as = 10-3.

We conducted simple simulations to verify the guarantee of QoS, i.e., the cell loss probability (CLP). The fGn with Hurst parameter 0.9, buffer size 1000 cells, and (u, v) = (1000, 100) is used as the traffic source. The allocated bandwidth is obtained from (5.27). The simulation results are summarized in Table 5.2 where the first column is the requested CLP ranging from 10-3 to 10-9, and the second and the third column show the estimated CLP by simulation for selected CAC schemes. Clearly the proposed CAC (PCAC) ensures the QoS. However, with the other statistical allocation algorithm, Equivalent Capacity (EC), the QoS is violated mainly because of the use of an inappropriate traffic model which fails to capture the characteristics of self-similarity.






71





requested CLP estimated CLP with PCAC estimated CLP with EC
10-3 6.2,10-4 3.5*10-2 10-4 5.9*10-5 4.6,1010-5 3.7,10-6 7.1*10-4 10-6 8.1*10-7 1.3*10-5 10-7 7.4*10-8 9.4*10-5 10-s 1.7*10-9 5.4*10-6 10-9 1.4*10-10 7.9*10-6 Table 5.2: Guarantee of QoS










120
H=0.5
m o r - . . . . - - - - --- -- . . . . . . .
o- --- -----H.





0 ...... .. -------------40


0H =..9 . ...... ...


< ........... .. ......

101 102 10 10' 10' Buffer size


Figure 5.3: Number of admitted connections vs. buffer size






72





H=0.7

... .y.-- ---I- -I" -+

70






C I


10 10 10 10 10 10 10
Cell loss probability


Figure 5.4: Number of admitted connections vs. cell loss probability

5.7 Conclusion

In this Chapter, we proposed a preventive flow control scheme consisting of a Connection Admission Control (CAC) taking into account the Hurst parameter of incoming traffic sources. The CAC is implemented based on an upper bound of the cell loss probability for the self-similar queueing model with a fGn arrival process. The numerical results showed that with the proposed CAC, more connections can be accepted into the networks than with peak rate allocation CAC.

In the proposed CAC, each source has to submit the traffic parameters including [L, v, and H, and desired Cell Loss Ratio e. This CAC is, therefore, better suited for applications such as Video on Demand (VOD). On the other hand, for the CAC to be able to support real-time traffic, these parameters have to be computed on the fly. It poses a challenge especially for estimating the Hurst parameter which requires longer time interval to obtain more accurate estimation.















CHAPTER 6
PID FLOW CONTROL SCHEME FOR VBR TRAFFIC

In communication networks with large delay-bandwidth product, congestion could happen over shorter time scales than those at which end-to-end protocols such as congestion control schemes typically operate. In such cases, the congestion can dissipate rapidly before congestion feedback information returns to the source. Network designers, therefore, face a challenge. The bursty and cyclic nature of Variable Bit Rate (VBR) traffic creates another issue for transmission in ATM networks. To reach the dual goals of keeping cell loss rate low and network utilization high, we propose an adaptive rate-based flow control scheme for real-time VBR traffic in ATM networks.

The goal of the scheme is to minimize the impact of traffic overload in order to limit the cell loss rate to an acceptable range and also increase the network utilization. The proposed flow control scheme is based on predicting the evolution of buffer occupancy over time using a Proportional-plus-Integral-plus-Derivative (PID) controller and a linear predictor to adaptively update the optimum data emission rate at the transmitter. The adaptive policy attempts to keep the buffer occupancy for each virtual channel at a steady level and the simulation results show that the proposed scheme works effectively against network congestion. Along with the design of the new flow control scheme, we also develop a hierarchically structured testbed to measure network performance and explore various flow control schemes in ATM networks with diverse classes of incoming traffic.






73






74


6.1 Introduction

Asynchronous Transfer Mode (ATM) is widely considered to be the next generation of high speed internetworking technology mainly due to the scalable building blocks it provides to serve as a stable foundation for evolving public and private networks. With the advent of Broadband ISDN, ATM networks must support applications with diverse traffic characteristics and quality of service (QoS) requirements. Therefore, an accurate traffic characterization is of primary importance in implementing effective flow control strategies and making efficient use of network resources in ATM networks.

Network traffic is usually characterized by the tolerance of information delay or loss. For example, the traffic for Available Bit Rate (ABR), or best-effort service, demands superior Cell Loss Rate (CLR) but yields inferior performance in terms of Cell Transfer Delay (CTD) and Cell Delay Variation (CDV). On the other hand, realtime traffic such as that for Constant Bit Rate (CBR) service is distinguished by its stringent CTD and CDV requirements. Real-time traffic contrasts significantly from elastic or best-effort traffic by primarily supporting applications that have dedicated bandwidth. These applications, however, do have some ability to adapt to network conditions such as cell delay or cell losses but only within a margin of tolerance [59]. The other type of real-time traffic in addition to CBR traffic is Variable Bit Rate (VBR) traffic such as that generated in videoconferencing using compression. The characteristics of VBR video traffic depends primarily on two factors:

1. the nature of the video, namely whether there are frequent or rare scene changes;

2. the coding scheme employed such as, for example, MPEG, H.261 etc.

For teleconferencing and videotelephone applications, the CCITT H.261 standard specifies compression techniques at rates of px64 kilobits/second where p ranges from 1 to about 30. Another well-known compression algorithms for video is MPEG,






75


Motion Pictures Expert Group defined by ISO. Compared to H.261, the MPEG standard was designed for a higher range of rates and much better visual quality. Although MPEG video is suitable for a large number of multimedia applications including videoconferencing, distant learning, and video mail, it is not, however, intended to be broadcast television quality which signals at 10 45 Mbps.

An MPEG coder generates three types of frames: Intraframes (I) that use image compression scheme, Predicted (P), and Bidirectional (B) frames that use compression with motion compensation. While P frames are coded based on the previous frame, B frames are based on both a past and a future reference frame. Therefore, I frames take advantage of spatial locality, and P and B frames exploit temporal locality. The MPEG algorithm uses several parameters that, when varied, can lead to a significant change in the characteristics of a video source. Two of these parameters, 1q, the quantization level and fI, the interframe to intraframe ratio, are of particular interest as they can be used to control the video source via network feedback [60]. For example, if the network detects a trend of congestion, a feedback message could be sent to all video sources that would require them to decrease their sending rate. This can be accomplished by decreasing 1q at the expense of a temporary loss of quality. Similarly, if the cell loss rate of the network temporarily increases, then, by decreasing fi the frequency of intraframes in the MPEG coding sequence could be increased. Since intraframes halt error propagation, this action will ensure that the duration of error propagation is short.

Figure 6.1(a) shows the frame size over time elapsed of a typical MPEG compressed video with f, equal to 29 (GOP=30), obtained from a recorded sequence of TV advertisements. Figure 6.1(b) depicts the relative frequency of the frame size. The encoding pattern is as follows.


IBBBBBBBBBPBBBBBBBBBPBBBBBBBBBIBB...







76



(a) frame size 300


E 200






0
cn 100



100 200 300 400 500 frames
(b) probability density function
0.05

0.04

0.03
o
0.02

0.01


50 100 150 200 250 cells


Figure 6.1: MPEG frame size and its probability density function


It is noted that between two successive I frames, a P frame is followed by 9 consecutive B frame. As Figure 6.1(b) shows, the frame sizes of this MPEG video fall into roughly 3 groups, with B frames occupying the frame size of less than 100 cells, I frames larger than 200 cells, and P frames in between.




The rest of the paper is organized as follows. Section 2 describes a generic video transmission system with feedback. In Section 3 we present the proposed adaptive flow control scheme. Section 4 describes the network scenario and the ATM switch used in our simulation. Section 5 shows some numerical results, and in Section

6 we state our conclusions and future work.






77


6.2 Generic Video Transmission System with Feedback

Among all traffic classes in ATM networks, real-time video traffic poses a unique challenge. Since the required service is delay sensitive, the network must provide a resource reservation scheme to allocate network resources for each VBR video stream. However, the burstiness of VBR video traffic makes it especially difficult to determine the amount of resources required. On the one hand, if resources are reserved according to the Sustainable Cell Rate (SCR) of the VBR video sources, unacceptable delays and cell losses may result when the video is transmitted at the peak rate. On the other hand, if resources reservations are based on the Peak Cell Rate (PCR) of the video sources, the network could be underutilized most of the time. One strategy to best compromise both problems is to dynamically adjust the sending rate of bitstreams according to the network information obtained via a feedback path between the network and the transmitter.

Figure 6.2 depicts a generic video transmission system, where a video signal is fed to a video encoder. The encoded bitstream is produced by video compression codecs such as MPEG or H.261 with an attempt to compress the video stream by removing spatial and temporal redundancies. The rate controller is able to select the transmission rate by adjusting the quantization level according to feedback information. This feedback path, however, in some standard such as H.261, has an option to disable it. This enables users to maximize quality at the expense of bandwidth.

Several flow control strategies for managing overload of VBR video traffic are described as follows.

Integrated rate-based flow control: Wei and et al [61] proposed a joint encoder and channel rate control algorithm for ATM networks with leaky buckets as open-loop source flow control models to balance both issues of consistent video quality on the encoder side and bitstream smoothness for the statistical multiplexing gain (SMG) on the network side. The algorithm considers






78



Video sender Video receiver
r - - - --- r- - - - --SVideo Channel Channel Video
Encoder Encoder Network Decoder Decoder Digita I video out Digital g
video in Playout
I Rate Controller Buffer
I _- I Feedback
L -- --L


Figure 6.2: Generic Video Transmission System with Feedback

constraints imposed by the encoder and decoder buffers, the leaky bucket control, traffic smoothing, and rate control. The encoder rate control is separated into a sustainable-rate control and a unidirectional instantaneous-rate control.

Lee and et al. [62] suggested a modified explicit rate indication for congestion avoidance (MERICA) scheme for both VBR and CBR traffics in ATM networks. This modified scheme belongs to the class of closed-loop rate-based

congestion control scheme.

* Adaptive credit-based flow control: Goyal and et al [63] presented a receiveroriented adaptive credit-based flow control scheme to minimize the buffer requirement in the network while guaranteeing that cells of VBR encoded video flows will not be lost. With the proposed bandwidth estimation techniques which exploit the structure of the video traffic, it also minimizes the end-to-end

delay and jitter for VBR encoded video streams.

* Data recovery: Lost data may be recovered using temporal and spatial interpolation or forward error recovery methods. Hu and Lin [64] suggested a forward error control scheme for MPEG video to recover lost cells in ATM networks.

The original MPEG streams are split into n substreams and the substreams are






79


skewed by a fixed amount of time from each other. The substreams are transmitted with the forward error correction code providing a deterministic bound on the delay and having the implementation simplicity needed for high-speed

video applications.

Prioritized transmission: Ismail and et al [65] proposed a priority scheme for

MPEG video which is implemented in a software encoder that produces a proportionate traffic in both (i.e., high and low) priority partitions for all three frame types used in MPEG. An ATM multiplexer with a pushout buffer scheme is implemented to provide priority scheduling at the multiplexer for the two priority partitions.

Traffic shaping and smoothing: Burstiness of encoded traffic may be removed

by smoothing it over an interval. Here one has to make a trade-off between cell losses due to burstiness and allowed delay. Li and et al [66] suggested a new flow control function called "time-driven" priority, which is an internal traffic shaping mechanism supporting CBR with deterministic guarantees, and VBR with statistical multiplexing. The mechanism does not require the identification and separation of packet flows of different real-time sessions/connections inside

the network.

Admission control: This mechanism can be used to accept or reject new connections at the connection set-up phase. ATM networks require traffic characteristics information such as CLR, PCR, SCR, and CDV for VBR traffic to

guarantee the QoS of a connection.

6.3 Proposed Flow Control Scheme

The proposed flow control scheme is comparable to a classical tracking control system, in which appropriate feedback is applied to have the system output y[n] track the input r[n] as Figure 6.3(a) shows. A compensator taking the difference between






80




(a)

r[n] + Compensator Plant y[n]






(b)

Controller Network bn






Figure 6.3: Generic tracking system and its application in networks

the reference r[n] and output y[n] as a trigger generates a dynamical signal entering the plant. The compensator must be well designed so y[n] can track r[n] promptly. One of the most well-known compensators is probably the so-called Proportionalplus-Integral-plus-Derivative (PID) controller. Figure 6.3(b) shows the scenario that applies the tracking system to networks. In this chapter, we exploit this system theoretic methodology in our proposed flow control scheme.



Ri,




r 64 Predictor bn



Figure 6.4: PID flow controller






81


In the proposed flow control scheme whose block diagram shown in Figure 6.4, R., the maximum sending rate of the nth frame is determined by the equation,


R, = Ri, [KP (-b*)+K ( b 1 b*) + KD 'n- bn-1) Ri (6.1) i=n-l+l

where Rin is the sending rate for the original bitstream of the nth frame without feedback, bn the buffer occupancy at the performance bottleneck at the beginning of the transmission of the nth frame, bn the estimated value of bn, b* the desired buffer occupancy, and 1 a non-negative integer.

In Equation 6.1, the actual sending rate Rn is updated once per frame interval, based on the buffer occupancy obtained from the feedback. The goal of the control action is to keep the buffer occupancy level close to the desired value b*. The sending rate Rn is adjusted by the controller so the buffer occupancy is driven to b* promptly. The estimated buffer occupancy b is obtained by using a linear predictor described by the following equation,

k-1
br = bn-k + [ R -i (k 1) p] tf (6.2) i=1
where p is a constant service rate at the buffer of interest, b,-k the latest reported buffer occupancy from feedback at the beginning of the transmission of the nthframe, and tf is the frame interval, i.e. the inverse value of the frame rate.

Let en denote the value of the error signal corresponding to the nth frame, that is e, = bn b*, Ae the difference of two consecutive error signals, and I(en, 1) the cumulative error signals in 1 frames. We can then rewrite Equation 6.1 as follows.



R, = Ri, (Kp, en + KI (en, 1) + KD Aen) Rin (6.3) Since @(en, 1) = e,, without the loss of generality when I > 1 Equation 6.3 can be further simplified to






82




Rn = Ri, (KI (e, 1) + KD Aen) Rin (6.4) Clearly we can see that, from (6.4), the instantaneous sending rate of the nth frame is determined by Kd, a derivative factor, KI, an integral factor, and I regarding cumulative error signals. Further, the values K, and KD control how rapidly the buffer occupancy approaches the desired buffer size.

6.4 Network Scenario for Simulations

Figure 6.5 shows the network scenario used in our simulations where two CPS-100 ATM switches, CPS 100-A and CPS 100-B, are connected to each other with a DS-3 trunk link at a speed of approximately 45 Mbps. The propagation delay between the two switches is 16.5 ms. There are four CPEs connected to each switch at both sides of the network. We model the data flows of the network such that they traverse from sources to corresponding destinations while the feedback transmitted in the reverse direction. All access ports are at the same speed as that of trunk ports.

We assume a one-to-one correspondence between sources and destinations, that is, each CPE at the left side of the network makes two VCs, one for voice and the other for video, to the corresponding CPE at the right. The simplified functional model of the CPE is shown in Figure 6.6. Due to the stringent QoS requirements of CBR service, the VCs for voice are given higher service and loss priority than those of video. Cells from the VCs with same priority will be served on a FIFO basis. As all traffic aggregates at the outgoing trunk port of S1, the egress buffer memory and the outgoing link bandwidth of the port become performance bottlenecks where congestion may occur when too many bursts arrive in a short period. The resource management (RM) cells carrying buffer information at the egress adapter are generated periodically and are sent back to all video sources by connections in the






83





TX-1 Rcv-1 TX-3 Rcv-3



CPS100-A CPS100-B TX-5 Rcv-6 TX-8 Rcv-8



Figure 6.5: Network Scenario reverse direction. Due to the balanced network topology, similar network behavior at each CPE is expected from simulations.

6.5 Numerical Results and Discussions

To evaluate the performance of the proposed PID flow control scheme under a congested network condition, each VC for voice generates traffic contributing 20% of the bandwidth of the trunk link, while each VC for video is contending for the





To CPS 100-A
CBR Flow Controller

Lk -9jFrom CPS 100-A VBR
VC for CBR
MEOVC for VBR
VC for Feedback


Figure 6.6: Functional model of CPE






84


remaining 20% of link speed, which is about 9 Mbps or 21,227 cells/sec. The trace of the video used in the simulation is shown in Figure 6.1. The frame rate of the video, 1/tf, is 30 frames/sec. In addition, the PCR and SCR is 7,700 cells/sec and 1,700cells/sec, respectively. The normalized network utilization rate PN is obtained by estimating the ratio of average VBR bandwidth to the overall bandwidth available to VBR. To simplify our simulation efforts and put the emphasis on the impact of K1 and KD, the value of 1 in Equation 6.4 is set to be 5.

Figure 6.7 and Figure 6.8 show the simulation results where the proposed flow control scheme is disabled. Figure 6.7(a) and Figure 6.7(b) show the frame size transmitted at CPE-T1 and that received at CPE-D1, respectively. The CLR (Cell Loss Rate) defined by the ratio of NT, the total number of cells transmitted, to NL, the total number of cell lost, can be obtain from Figure 6.8 (a) and Figure 6.8 (b). The CLR in this case is 0.046 (NT = 19285, NL = 887). The CLR is clearly less than 5% without using the rate controller, but PN, the normalized network utilization rate, is as low as 30% which means the bandwidth available to VBR traffic is unused 70% of the time.

Figure 6.9 and Figure 6.10 show the results for the same statistics as those of Figure 6.7 and Figure 6.8 when the proposed flow control scheme is used with (KI, KD) = (0.2, 0.05). Notice that, with more cells transmitted (and better video quality) than the case of no feedback, the CLR increases to 0.072 (NT = 20800, NL = 1488), but PN is improved substantially to 43%. The results under different values of (KI, KD) are listed in Table 6.1. Both CLR and PN increase as the values of KI, KD increase.

Another opportunity to take advantage of the proposed flow control scheme and make the most use of available bandwidth is to aggregate more VBR streams. Let Nv. denote the number of VCs for video at each CPE. The CLR and PN corresponding to different value of Nvb, are shown in Table 6.2. As Nvbr increases, more bursty video






85




(a)
300 250
200

a) 150

100
50

0 2 4 6 8 10 time (sec)
(b)
300 250
200
a) 150
100 50

0 2 4 6 8 10 time (sec)



Figure 6.7: Transmitted and received frame size with no feedback control


(KI, KD) CLR PN (0.00, 0.00) 0.046 0.301 (0.20, 0.00) 0.073 0.427 (0.20, 0.05) 0.072 0.433 (0.20, 0.15) 0.081 0.456 (0.00, 0.15) 0.053 0.358 (0.10, 0.15) 0.062 0.405 (0.15, 0.15) 0.068 0.418 Table 6.1: CLR and normalized utilization rate under various (KI, KD)






86











x 104 (a)
2.5

2

1.5

1

0.5


0 2 4 6 8 10 time (sec)
(b)
2000 1500

0 1000

500


0 2 4 6 8 10 time (sec)



Figure 6.8: Cumulative number of cells sent and lost with no feedback control






87











(a)
300 250 200

S150


50

0 2 4 6 8 10 time (sec)
(b)
300 250 200

a 150


50

0 2 4 6 8 10 time (sec)



Figure 6.9: Transmitted and received frame size with feedback control






88











x 104 (a)
2.5

2

S1.5 1


0.5


0 2 4 6 8 10 time (sec)
(b)
2000 1500

S1000

500


0 2 4 6 8 10 time (sec)



Figure 6.10: Cumulative number of cells sent and lost with feedback control






89

Nvbr CLR PN
1 0.072 0.433 2 0.082 0.521 3 0.102 0.604 4 0.167 0.892 5 0.304 0.954 Table 6.2: CLR and normalized utilization rate vs. number of VBR VCs

streams get aggregated on the trunk link. Consequently, both CLR and PN increase as predicted.

6.6 Summary

In this chapter, we presented a closed-loop rate-based flow control scheme for real-time VBR traffic in ATM networks using a methodology based on the classical PID controller. The primary goal of the flow control scheme is to limit the CLR to an acceptable range and substantially increase the network utilization rate by adjusting the sending rate of the video stream at the transmitter dynamically based on the network condition obtained from the feedback. The simulation results show that the proposed scheme works quite well.














CHAPTER 7
INTEGRATED SERVICES PROTOCOLS AND TRAFFIC ANALYSIS

The Internet has proven to be the best solution for global sharing of information. Through the Internet, one can access information from world wide web (WWW) and exchange electronic mail, as well as share applications with others from any continent on earth. The success of these end-to-end communications is essentially achieved by two transport layer protocols in the Internet protocol suite, namely, Transport Control Protocol (TCP) and User Datagram Protocol (UDP), where TCP provides a connection-oriented, reliable flow between two hosts and UDP provides a connectionless, unreliable datagram service.

Currently one class of service in the Internet exists normally referred to as "best effort" whose traffic is characterized by first come first serve scheduling at each node in the network. Best effort service has worked extremely well for traffic without time constraint. However, for real-time traffic, such as voice and video, TCP/UDP has performed well only when networks are underloaded. In order to provide guaranteed quality of service (QoS) for real-time traffic ceaselessly, new classes of service and new protocols are being introduced on the Internet while retaining the existing best effort service. Three emerging technologies that will be able to support the needed Internet multimedia services are "Resource Reservation Protocol (RSVP)", "Real-Time Transport Protocol (RTP)", and "IP Multicast Protocol."

Network traces have been widely considered as one of the most efficient ways to study network dynamics, usage characteristics, and growth patterns. They are of extreme importance to trace-driven simulations for tools like OPNET. In this Chapter, we describe the three real-time oriented protocols mentioned above in Session 1.


90




Full Text
ACKNOWLEDGEMENTS
I would like to thank the members of my supervisory committee for their
help and guidance throughout this work. I would like to express special thanks to
my committee chairman, Dr. Latchman, for his invaluable instruction and advice
through this research. Also, special thanks to Dr. Chow, Dr. Couch, Dr. Taylor and
Dr. Newman-Wolfe for their helpful suggestions and comments on this dissertation.
I am grateful to the many colleagues and friends who encouraged me to go on
for my graduate career. My special thanks go to Dr. Shang-Yi (Debbie) Lu, whose
advice and expertise have been invaluable to this work. I am indebted to my wife,
Hui-hsien, and my son, Kevin, for accompanying me through the years without any
complaint. Finally, I would like to thank my parents and my sisters for their full
support and understanding during my long and fruitful education.
n


process. In this Ph.D. research, we investigate the properties of self-similar processes
in great detail. We present a queueing model with a fractional Gaussian noise arrival
process and deterministic service time. This queueing model is radically distinct from
conventional queueing models whose arrival process does not take into account the
self-similarity of incoming traffic. Our simulation results show that the latter leads to
a substantial amount of inaccuracy in terms of network performance. Based on the
queueing model, we propose a preventive flow control scheme suited for self-similar
traffic using a Connection Admission Control (CAC). This CAC takes into account
the Hurst parameter of traffic sources. The numerical results show that the proposed
CAC ensures the QoS and achieves more admitted connections than the peak rate
allocation CAC does.
In this research, we also propose a rate-based flow control scheme suited for
real-time traffic. The main goal of the flow control scheme is to increase the network
utilization with a margin of degradation in cell loss ratio. The proposed flow control
scheme predicts the evolution of buffer occupancy over time using a linear predictor.
We use a Proportional-plus-Integral-plus-Derivative (PID) controller to update the
optimum sending rate at the transmitter dynamically. The adaptive policy attempts
to keep the buffer occupancy for each virtual channel at a steady level, and the
simulation results show that the proposed scheme works effectively against network
congestion.
x


72
Figure 5.4: Number of admitted connections vs. cell loss probability
5.7 Conclusion
In this Chapter, we proposed a preventive flow control scheme consisting of
a Connection Admission Control (CAC) taking into account the Hurst parameter of
incoming traffic sources. The CAC is implemented based on an upper bound of the
cell loss probability for the self-similar queueing model with a fGn arrival process.
The numerical results showed that with the proposed CAC, more connections can be
accepted into the networks than with peak rate allocation CAC.
In the proposed CAC, each source has to submit the traffic parameters in
cluding p, v, and H, and desired Cell Loss Ratio e. This CAC is, therefore, better
suited for applications such as Video on Demand (VOD). On the other hand, for the
CAC to be able to support real-time traffic, these parameters have to be computed
on the fly. It poses a challenge especially for estimating the Hurst parameter which
requires longer time interval to obtain more accurate estimation.


CHAPTER 3
TRAFFIC MODELLING AND SELF-SIMILAR PROCESSES
A recently observed phenomenon related to traffic in Broadband Integrated
Service Digital Networks (B-ISDN) is the self-similarity or burstiness exhibited
by key services such as compressed video [3], file transfer [5], and WWW [5]. Two
main sources of burstiness are due to the shapes of the marginal distribution as well
as the autocorrelation function of packet arrival rate or interarrival times [22].
When modeling network traffic, packet and connection arrivals are often as
sumed to be Poisson processes because such processes have attractive theoretical
properties. A number of studies have shown, however, that for both local-area and
wide-area network traffic, the distribution of packets inter arrival time clearly differs
from exponential [23, 24, 25, 26, 27]. Recent works argue convincingly that LAN
traffic is much better modeled using statistically self-similar processes, which have
significantly different theoretical properties than Poisson processes [28, 2]. An appro
priate traffic modeling which reflects the burstiness or self-similarity is, hence, critical
to the success of performance evaluation.
The remainder of this chapter is organized as follows. Commonly used Poisson
and Poisson-based traffic models are presented in Section 1. The self-similar processes
and their properties are given in Section 2. Section 3 describes the modeling of Self
similar processes, and finally several popular self-similar processes generators are
described in Section 4.
26


77
6.2 Generic Video Transmission System with Feedback
Among all traffic classes in ATM networks, real-time video traffic poses a
unique challenge. Since the required service is delay sensitive, the network must pro
vide a resource reservation scheme to allocate network resources for each VBR video
stream. However, the burstiness of VBR video traffic makes it especially difficult
to determine the amount of resources required. On the one hand, if resources are
reserved according to the Sustainable Cell Rate (SCR) of the VBR video sources, un
acceptable delays and cell losses may result when the video is transmitted at the peak
rate. On the other hand, if resources reservations are based on the Peak Cell Rate
(PCR) of the video sources, the network could be underutilized most of the time.
One strategy to best compromise both problems is to dynamically adjust the sending
rate of bitstreams according to the network information obtained via a feedback path
between the network and the transmitter.
Figure 6.2 depicts a generic video transmission system, where a video signal
is fed to a video encoder. The encoded bitstream is produced by video compression
codecs such as MPEG or H.261 with an attempt to compress the video stream by
removing spatial and temporal redundancies. The rate controller is able to select the
transmission rate by adjusting the quantization level according to feedback informa
tion. This feedback path, however, in some standard such as H.261, has an option to
disable it. This enables users to maximize quality at the expense of bandwidth.
Several flow control strategies for managing overload of VBR video traffic are
described as follows.
Integrated rate-based flow control: Wei and et al [61] proposed a joint en
coder and channel rate control algorithm for ATM networks with leaky buckets
as open-loop source flow control models to balance both issues of consistent
video quality on the encoder side and bitstream smoothness for the statisti
cal multiplexing gain (SMG) on the network side. The algorithm considers


CHAPTER 4
CONGESTION AND FLOW CONTROL
The B-ISDN, which is based on the ATM technique, is designed to transport
a wide variety of traffic classes satisfying a range of transfer capacity needs and
network performance objectives. The underachievement, if any, is usually caused by
network congestion which could happen whenever the overall instantaneous input
rate is greater than the link capacity for a certain period of time, i.e.
£A I>C (4.1)
where A/ represents the instantaneous input rate of an active source, and C the
link capacity. To be more general, congestion in ATM networks can be caused by
statistical fluctuation of traffic flows or network failures [20]. In light of the diverse
traffic characteristics in ATM networks, the design of a scheme to adequately control
the data flow into ATM networks so as to avoid congestion becomes a critical factor
for the success of B-ISDN. Aside from the primary role of congestion and flow control
schemes to protect the network and the user in order to achieve network performance
objectives, an additional role is to optimize the use of the network resources so as to
achieve better utilization efficiency.
4.1 General Framework for Congestion Management and Control
For a congestion management and control scheme to be able to support a set of
QoS classes sufficient for all foreseeable services and maintain minimum network and
end-system complexity while maximize network utilization, the following functions
46


9
transmission over unshielded twisted pair (UTP). The CCITT/ITU-T has concerned
itself mainly with standards suitable for use with SDH (Synchronotis Digital Hierar
chy) /SONET transmission networks and has defined standards for optical fiber at
speed of 155 Mbps, 622 Mbps, and 2.4 Gbps. Some other relevant CCITT/ITU-T
standards for the Physical Layer are G.703, G.707, G.708, G.709 and 1.432 [20, 16].
Proposals to transmit ATM cells at 2 Mbps, 1.5 Mbps, and other speeds also exist.
2.1.3 ATM Layer
The ATM layer is specified in ITU-T Recommendation 1.361 to provide for
transparent transport of the data between AAL entities. This transfer takes place
on a pre-established ATM connection according to a transfer contract consisting of a
Quality of Service (QoS) class, a vector of traffic parameters. The ATM layer takes
streams of 48-byte cell payloads as its input from higher layers, performs cell header
generation, and passes cells to the TC sublayer of the physical layer such that order
is preserved within virtual channels (VCs). Two levels of VCs can be supported at
the user-network interface (UNI) that connects customer premises equipment (CPE)
and ATM switches:
1. A point-to-point, or point-to-multipoint, Virtual Channel Connection (VCC)
consisting of a single connection established between two ATM VCC end points.
2. A point-to-point, or point-to-multipoint, Virtual Path Connection (VPC) con
sisting of a bundle of connections established between two ATM VPC end
points.
The ATM cell structure, in Figure 2.2, shows the first five bytes are allocated
to the header of a fixed 53-byte cell. The entire header is protected by a l-byte
Header Error Check (FEC) field. We noticed that there is no retransmission of lost
or corrupted data preformed by this layer. An access flow control mechanism may


76
(a) frame size
50 100 150 200 250
cells
Figure 6.1: MPEG frame size and its probability density function
It is noted that between two successive I frames, a P frame is followed by 9 consecutive
B frame. As Figure 6.1(b) shows, the frame sizes of this MPEG video fall into roughly
3 groups, with B frames occupying the frame size of less than 100 cells, I frames larger
than 200 cells, and P frames in between.
The rest of the paper is organized as follows. Section 2 describes a generic
video transmission system with feedback. In Section 3 we present the proposed
adaptive flow control scheme. Section 4 describes the network scenario and the ATM
switch used in our simulation. Section 5 shows some numerical results, and in Section
6 we state our conclusions and future work.


3.2.3 Self-similarity and Hurst Effect 36
3.2.4 Self-similarity and Slowly Decaying Variances 36
3.3 Modeling of Self-Similarity 37
3.3.1 Fractional Brownian Motion 37
3.3.2 Fractional Gaussian Noise 38
3.3.3 Fractional ARIMA(p,d,q) Processes 39
3.3.4 Self-Similarity Through Aggregation 41
3.4 Self-similar Traffic Trace Generators 42
3.4.1 Random Midpoint Displacement Algorithm 42
3.4.2 Aggregation of Renewal Processes 44
3.4.3 M/G/oo with Heavy-tailed Distributed Service Time ... 44
3.4.4 ARIMA 44
3.4.5 Approximation of Power Spectrum 45
4 CONGESTION AND FLOW CONTROL 46
4.1 General Framework for Congestion Management and Control . 46
4.2 Flow Control Schemes with Feedback 49
4.2.1 Credit-Based Scheme 49
4.2.2 Rate-Based Scheme 50
4.3 Impropriety of Poisson-Based Traffic Models 52
4.3.1 Network Configuration 52
4.3.2 Effective Throughput Comparisons 53
4.4 Classical Control Theory and Congestion Control 55
5 PREVENTIVE FLOW CONTROL SCHEME FOR SELF-SIMILAR TRAF
FIC 58
5.1 Introduction 59
5.2 Peak Rate Allocation 60
5.3 Statistical Allocation 61
5.4 Cell Loss Probability in Self-similar Queuing Models 61
5.5 Proposed Call Admission Control Algorithm 67
5.6 Numerical Results and Discussions 67
5.7 Conclusion 72
6 PID FLOW CONTROL SCHEME FOR VBR TRAFFIC 73
6.1 Introduction 74
6.2 Generic Video Transmission System with Feedback 77
6.3 Proposed Flow Control Scheme 79
6.4 Network Scenario for Simulations 82
6.5 Numerical Results and Discussions 83
6.6 Summary 89
7 INTEGRATED SERVICES PROTOCOLS AND TRAFFIC ANALYSIS 90
7.1 Protocols for Integrated Services 91
7.1.1 Resource Reservation Protocol 91
7.1.2 Real Time Transport Protocol 93
7.1.3 IP Multicast Protocol and Multicast Backbone 94
7.2 Traffic Measurements and Analysis 96
7.2.1 Network topology and Network tool 96
7.2.2 Traffic Measurements and Analysis 100
IV


CHAPTER 1
INTRODUCTION
Asynchronous Transfer Mode (ATM) is one of the most promising and widely
recognized solutions for the high speed, multimedia networks of next generation. The
anticipation is mainly due to the scalable building blocks ATM provides to serve as
a stable foundation for evolving public and private networks. ATM is capable of
handling many kinds of information including voice, data, image and video in an
integrated manner and was accepted as the transfer mode for Broadband Integrated
Services Digital Networks (B-ISDN) by the International Telecommunications Union
(ITU) in 1988. With the advent of B-ISDN, ATM must support applications with
diverse traffic characteristics and quality of service (QoS) requirements. An accu
rate traffic characterization and modeling is, therefore, of primary importance in
implementing effective flow control strategies and in making efficient use of network
resources in ATM networks.
1.1 Motivations and Objectives
ATM has been effectively making inroads in the world-wide networks and has
made substantial progress in developing key standards regarding signaling, traffic
management and LAN emulation. However, there still exist some challenges that we
need to address in order to facilitate the design and optimal performance of functional
ATM networks. This Ph.D. research presents two of these key issues, namely an
appropriate characterization and modeling of ATM traffic and effective congestion
management schemes suitable for the self-similar or fractal-like traffic expected on
ATM networks.
1


Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
EFFICIENT FLOW AND CONGESTION CONTROL
FOR
SELF-SIMILAR TRAFFIC
IN
ATM NETWORKS
By
Wen-Yen Fn
May 1998
Chairman: Dr. Haniph A. Latchman
Major Department: Electrical and Computer Engineering
Asynchronous Transfer Mode (ATM) has emerged as one of the most promis
ing solutions for the next generation networks since being adopted as the transfer
mode of broadband integrated services digital networks (B-ISDN). With the advent
of the high speed, multimedia networks, ATM must support applications with diverse
traffic characteristics and quality of service (QoS) requirements. An accurate traffic
characterization and modeling is, therefore, of primary importance in implementing
effective flow control strategies and in making efficient use of network resources in
ATM networks.
Recent studies have verified self-similar or fractal-like behaviors of the traffic
over local and wide area networks, which implies that it is more appropriate to
model this traffic using a self-similar process than using a traditional Poisson-based
IX


TABLE OF CONTENTS
ACKNOWLEDGEMENTS ii
LIST OF TABLES vi
LIST OF FIGURES vii
ABSTRACT x
CHAPTERS
1 INTRODUCTION 1
1.1 Motivations and Objectives 1
1.2 Dissertation Outline 3
2 BACKGROUND 6
2.1 Asynchronous Transfer Mode 6
2.1.1 ATM Layered Structure 7
2.1.2 Physical Layer 8
2.1.3 ATM Layer 9
2.1.4 ATM Adaptation Layer 11
2.2 Quality of Service 14
2.2.1 Accuracy 15
2.2.2 Speed 16
2.3 Traffic Characteristics 16
2.3.1 CBR Traffic 17
2.3.2 VBR Traffic 17
2.3.3 ABR Traffic 18
2.4 CPS100 Example of ATM Switching Design 19
2.4.1 Interface Modules 20
2.4.2 Advantages and Limitations of CPS-100 Switch 24
3 TRAFFIC MODELLING AND SELF-SIMILAR PROCESSES .... 26
3.1 Poisson-Based Traffic Models 27
3.1.1 Poisson Process Models 27
3.1.2 Markov Process Models 28
3.1.3 Markov-Modulated Process Models 29
3.2 Self-similar Processes 30
3.2.1 Definitions 32
3.2.2 Self-similarity and Long-Range Dependence 34
iii


33
where O < H < 1 and S(t) is slowly varying at infinity, namely lim^oo S(tx)/S(t) =
1, for all x > 0. Denote Wm) the new process obtained by averaging the original
series X in non-overlapping sub blocks of size m. That is:
1 m 1
XW(k) = Y, Xkm-i (3.8)
171 i=0
Note that for each m, defines a WSS random process.
Definition 2 Denote r^m\T) the autocorrelation function of X^m\ The process X(t)
is exactly second-order self-similar [1] with Hurst parameter H if
r^(r) = r(r), for all m = 1,2,... (3.9)
In other words, X is exactly second-order self-similar if the aggregated pro
cesses are indistinguishable from X with respect to their first and second order prop
erties. With the relaxation of (3.9), we arrive at the following definition.
Definition 3 X is asymptotically self-similar with Hurst parameter H if
r^m\r) = 1/2 [(r + l)2H + (r \)2H 2r2H], as m > oo. (3.10)
Specifically, for large m, the aggregated process X^ has a fixed autocorrela
tion structure determined only by H.
The most striking feature of self-similarity is that the correlation functions of
the aggregated processes do not decay to 0 as m > oo. This feature is in contrast
to traditional models, all of which have the property that the correlation function of
their aggregated processes decline as m oo. i.e.,
r(m)(r) = 0, as 77i > oo, for r = 1,2, 3,.. (3-11)
Two of the most well-known self-similar processes are fractional Gaussian noise
(fGn) and its incremental version, fractional Brownian motion (fBm), which will be
discussed in detail later in this chapter.


94
number, and these fields can be used by a destination host to reconstruct the tempo
ral information of RTP packet streams. In addition, the delivery or data transport is
enhanced by another protocol, RTP Control Protocol (RTCP), designed to provide
minimal control and identification. Since RTCP appears as just another option field
within a RTP packet, RTP can be used without RTCP if desired.
Applications typically run RTP on top of UDP as part of the transport layer
protocol, as shown in Figure 7.7. In practice, RTP is usually implemented within the
application. To set up an RTP session, the application defines a particular pair of
destination transport addresses containing one network address plus a pair of ports
for RTP and RTCP. For example, in a multicast audio/video conference as shown in
Figure 7.5, it is a common situation to have members of the working group join and
leave during the session. It is therefore desirable to know who is participating at any
moment and how well they are receiving the data. For this purpose, each participant
in the conference periodically multicasts a reception report plus the name of its user
on the RTCP port. The reception report indicates how well the current data is being
received and may be used to control adaptive encoding as well as enable a recipient
to select whether or not to receive a particular medium.
Despite that the Internet was its original design target, RTP does not assume
anything about the underlying transport protocol, so it can be used over unicast as
well as multicast. For example, test runs of RTP transmissions over ATM/AAL5 are
in progress in both Internet Engineering Task Force (IETF) and ATM Forum.
7.1.3 IP Multicast Protocol and Multicast Backbone
IP Multicast Protocol facilitates distributed applications to achieve time-
critical communications such as videoconferencing over wide area IP networks [71, 72],
The IP Multicast routers, typically referred to as mrouters, take the responsibility
of distributing and replicating multicast data streams to their destinations as opposed


49
or end-to-end feedback. Hop-by-hop feedback requires each intermediate switching
node to send the information to its previous stage and is, therefore, more suited for
services which feature short-term overloads. On the other hand, end-to-end feedback
is more effective for services with long-term overloads because of its capability to
react to the feedback in a more promptly manner.
4.2 Flow Control Schemes with Feedback
Most congestion control schemes for ATM networks consist of adjusting the
input rates to match the available link rate in attempt to improve the network uti
lization efficiency via feedback. Flow control schemes using feedback are spearheaded
by two approaches: rate-based, and credit-based flow control scheme. After a series
of intensive debates in ATM forum, the former was finally selected as the standard of
the flow controller in B-ISDN mainly thanks to its higher cost effectiveness compared
to the latter.
4.2.1 Credit-Based Scheme
This is one of the two leading approaches and also the first one to be proposed,
analyzed, and implemented. The approach consists of per-link, per-VC (Virtual
Channel), and window flow control. Each link consists of a sender node and a receiver
node. Each node maintains a separate queue for each VC. The receiver monitors
queue length of each VC and determines the number of cells that the sender can
transmit on that VC. This number is called credit. The sender transmits only as
many cells as allowed by the credit. In the case of a single active VC, the credit must
be large enough to allow the whole link to be full at all times. In other words:
Credit > Link Cell Rate x Link Round Trip Delay, (4.2)
where one can compute the link cell rate by dividing the link bandwidth in Mbps by
the cell size in bits.


LIST OF FIGURES
2.1 B-ISDN Protocol reference model 8
2.2 ATM Cell Format at UNI and NNI 10
2.3 TV Comercial 18
2.4 TV News 19
2.5 Functional model of the CPS-100 switch 21
2.6 Adapter Interface of a Shared Medium/Output Buffering ATM Switch 22
3.1 2-state Markov chain 29
3.2 Self-similarity of WWW traffic 31
3.3 Autocorrelation function and long-range dependence 35
3.4 Fractional Brownian motion and fractional Gaussian noise 43
4.1 Network scenario for Poisson vs. Self-similar traffic models 53
4.2 Variance-time plot for synthesized sample paths 54
4.3 Performance comparison of different flow control schemes 55
4.4 Performance comparison of rate-based scheme I 56
4.5 Performance comparison of rate-based scheme II 56
4.6 Generic Feedback Control System 57
5.1 Cell loss probability vs. buffer size 68
5.2 Estimated CLP vs. genuine Hurst parameters 70
5.3 Number of admitted connections vs. buffer size 71
5.4 Number of admitted connections vs. cell loss probability 72
vii


42
3.4 Self-similar Traffic Trace Generators
While most network management tools such as HPs OpenView and IBMs
Net View have built-in capabilities to monitor and gather trace data, not everyone
has access to such tools. One of the most popular monitoring tools designed as user
commands is tcpdump, which is useful to get an overview of what type of or how
many packets are on a network for a given time period. The problem is that not many
people feel comfortable reading and recording them, mostly due to the concerns of
individual privacy or network security. However, to do simulation studies, a lot of
trace data is needed, and gathering and storing this data not only is time consuming
but also requires huge storage space. What is needed is a fast method of generating
synthetic traffic traces for use in simulations. Currently, a number of approaches to
generate self-similar sample paths are available, and we introduce several popular
ones in the following sections.
3.4.1 Random Midpoint Displacement Algorithm
The most well-known algorithm for generating fBm is the Random Mid
point Displacement (RMD) algorithm. RMD works by progressively subdividing
an interval over which to generate the sample path. At each division, a random
displacement drawn from a Gaussian distribution is used to determine the value of
the sample path at the midpoint of the subinterval. Self-similarity comes about by
appropriately scaling the variance of the displacement. A Matlab script for RMD is
included in Appendix B.l. Examples of the resultant fBm and fGn using this script
are depicted in Figure 3.4, where fBm and fGn are shown in the first and the second
column, respectively. In this figure, various Hurst parameters with a value of 0.1, 0.5,
and 0.9 are shown in the first row, the second row and the third row, respectively.
Note that fGn is simply the increments of fBm. Also note that negative dependence
and LRD are observed as H=0.1 and H=0.9, respectively.


APPENDIX A
SCRIPTS FOR GENERATING SELF-SIMILAR TRAFFIC
A.l Random Mid-point Displacement Algorithm
7. Random Midpint Displacement Algorithm
% x: intended Hurst parameter
7, y: number of point to generate
function [a,b,t.scale]=brown(x,y)
h=x;
maxlevel=floor(log(y)/log(2)) ;
start=l;
length=2~maxlevel+l;
t=l/sqrt(12(2*h-2));
a=zeros(length,1);
b=zeros(length,1);
scale=zeros(maxlevel,1) ;
/.scale factor
7.
for i= l:maxlevel,
scale(i)=t*0.5~(h*(i1));
end
/.subdivisin
a(1ength)=randn;
b(length)=a(length);
level=l;
110


81
In the proposed flow control scheme whose block diagram shown in Figure 6.4,
Rn, the maximum sending rate of the nth frame is determined by the equation,
Rn = Rin [Kp (bn -b*) + Kj-( bi-l-b^ + Ko-ibn-bn-^-Rin (6.1)
i=nl+1
where Rin is the sending rate for the original bitstream of the nth frame without
feedback, bn the buffer occupancy at the performance bottleneck at the beginning of
the transmission of the nth frame, bn the estimated value of bn, b* the desired buffer
occupancy, and l a non-negative integer.
In Equation 6.1, the actual sending rate Rn is updated once per frame interval,
based on the buffer occupancy obtained from the feedback. The goal of the control
action is to keep the buffer occupancy level close to the desired value b*. The sending
rate Rn is adjusted by the controller so the buffer occupancy is driven to b* promptly.
The estimated buffer occupancy b is obtained by using a linear predictor described
by the following equation,
fc-i
bn = K-k + E/2n-i (k 1) p] tf (6.2)
i= 1
where g is a constant service rate at the buffer of interest, bn-k the latest reported
buffer occupancy from feedback at the beginning of the transmission of the nframe,
and tf is the frame interval, i.e. the inverse value of the frame rate.
Let en denote the value of the error signal corresponding to the nth frame, that
is en = bn b*, Aen the difference of two consecutive error signals, and cumulative error signals in l frames. We can then rewrite Equation 6.1 as follows.
Rn = Rin (Kp en + K¡ $(en, l) + KD Aen) Rin (6.3)
Since 1 Equation 6.3 can be
further simplified to


61
utilization rate is R-1. Consequently, PRA is best suited for CBR services including
PCM-encoded voice, uncompressed video and applications with low PCR such as
telemetry.
5.3 Statistical Allocation
In statistical allocation, bandwidth for a new connection is not allocated based
on PCR. In consequence, the sum of all peak rates may be greater than the link speed.
Statistical allocation makes economic sense when dealing with bursty sources. One
primary drawback of statistical allocation is its difficulty carrying out effectively,
mainly due to the difficulties in characterizing an arrival process.
An example of statistical allocation is the use of Equivalent Capacity [55]
defined as the service rate of the queue that corresponds to a cell loss probability
e. By using an interrupted fluid process as the traffic model, [55] shows that given
a finite buffer size K, peak rate R, MCR p, and the utilization rate p, the required
bandwidth to achieve cell loss probability e is
(5.1)
where a = ln(e)p(l p)R.
5.4 Cell Loss Probability in Self-similar Queuing Models
To evaluate the cell loss probability in an ATM connection, we studied the per
formance of a queueing system loaded by the composition of independent fractional
Gaussian noise (fGn) processes. We described a proposed diffusion approximation
for the number of arrivals in the interval (0,t], considering an heavy traffic renewal
processes below.
Proposition 2 In a queueing system with renewal arrival process with mean p, the
cumulative counts of arrival, a(t), can be approximated by
a(t) = pt + yfpB(i),
(5.2)


12
transform a sequence of cells into the byte stream used to transport digitized signals.
This is performed by the ATM Adaptation Layer (AAL), as illustrated in Figure 2.1.
The ATM adaptation layer (AAL) converts the real data stream into a cell stream
and vice versa. Given the wide variety of possible characteristics of the input data
streams, one would expect to find a variety of different adaptation layers. ITU Rec
ommendation 1.362 defines the basic principles and classification of AAL functions,
which are described as follows.
AAL Services Classes
Class A: constant bit-rate (CBR) service with end-to-end timing, connection-
oriented. The examples are fixed-rate video and the circuit emulation services
such as DSl or DS3 transport.
Class B: variable bit-rate (VBR) service with end-to-end timing, connection-
oriented. The examples are packetized voice and video.
Class C: variable bit-rate (VBR) service with no end-to-end timing, connection-
oriented
Class D: variable bit-rate (VBR) service with no end-to-end timing, connec
tionless
AAL Functions
AAL1 is designed to transport connection-oriented, CBR data stream in such
a way that clock information can be recovered at the receiving end (time trans
parency). AAL1 is, in effect, a virtual wire. Moreover, it can correct single-bit
errors in the payload and notifies lost cells or misordered cells.
AAL2 is specified to transport connection-oriented, VBR data streams in such
a way that timing information is recoverable at the receiving end. AAL2 has


30
used to model voice traffic sources, where the on state corresponds to talk spurt,
and the off state corresponds to silence.
3.2 Self-similar Processes
One of the most remarkable aspects about self-similarity in 90s is probably
that it has been consistently observed in recent studies and measurements in network
traffic [27, 5, 29]. Self-similarity is manifested in such a way that there is no natural
length for a burst across all time scales. In other words, a bursts consists of
bursty subperiods separated by less bursty subperiods. For instance, fractal images
in computer graphics and snowflakes are probably two of the most well-known objects
exhibiting self-similar or fractal-like phenomena where each small portion of them
can be viewed as a reduced-scale duplicate of the whole. Some aspects of self-
similarity, coined by Mandelbrot [30], also appears in hydrology, economics, and
communications and is related to so called 1/f noises. [31]
Self-similar traffic in networks behaves so differently from voice traffic, that
none of the traditional Poisson-based traffic models described in Session 3.1 is able to
capture the fractal behaviors. It is also shown that the generally accepted argument
that aggregate traffic becomes smoother as the number of traffic sources increase
has very little to do with the reality in Ethernet LANs [27]. On the contrary, the
burstiness of LAN traffic typically intensifies as the amount of active traffic increases.
Figure 3.2 illustrates the empirical trace collected from a WWW server at a
typical Internet service provider (ISP) in June 1997. The breakdowns of packet arrival
rate are plotted according to time scales ranging from 0.06 second to 60 seconds. One
can observe the similar appearance of burstiness across a wide range of time scales,
where the plot on the top corresponds to the complete trace with a resolution of 60
seconds, the second from the top an extraction of the first one (l/10th) but with ten
times finer resolution (6 seconds), and so on. It is evident that the burstiness persists
over different time-scales, and that the traffic is statistically self-similar.


99
(a) video (b) audio
fT-l LIST test run _
Wen-Yen Fu
\W lleten
fTST
f
.
! A Keep Audio
LBL Visa*/Audio Too/ v4 0*2 Menu |
Help 1 Quit |
Figure 7.6: Video and audio streams in a MBone session
Applications
A .
TELNET
FTP
WWW
SMTP
X11
NNTP
DNS
SNMP
SIN
DHCP
NFS
OIA
VAT
SDR
RJCP| RTp
TCP
UDP
IP
Figure 7.7: Network applications and TCP/IP protocol stack


112
zmirr=conj(fliplr(z));
zexpand=[0 z zmirr(2:n)];
x=real(ifft(zexpand));
plot(x);
ra
function [psd]=psd_fgn(h,lamda)
n=size(lamda);
d=-2*h-l;
dprime=-2*h;
A=2*sin(pi*h)*gamma(-d)*(ones(l ,n)-cos(lamda));
al=2*pi*ones(1,n)+lamda;
bl=2*pi*ones(l,n)-lamda;
a2=2*2*pi*ones(1,n)+lamda;
b2=2*2*pi*ones(1 ,n)-lamda;
a3=2*3*pi*ones(1,n)+lamda;
b3=2*3*pi*ones(1,n)-lamda;
a4=2*4*pi*ones(1,n)+lamda;
b4=2*4*pi*ones(1,n)-lamda;
B=al.~d+bl.~d+a2.~d+b2.~d+a3.d+b3.~d+...
(a3.~dprime + b3.~dprime + a4.~dprime + b4.~dprime)/8/pi/h;
psd=A.*[(abs(lamda)).~d+B];
/ end of PSD approximation


47
form a framework for managing and controlling congestion in ATM networks and
may be used in appropriate combinations [20].
Resource Management (RM) [14, 40, 41]: RM provides for allocating network
resources in order to separate traffic flow according to the QoS on traffic con
tract. Two critical resources in ATM networks are buffer space and trunk
bandwidth.
Connection (or Call) Admission Control (CAC) [42, 43, 44]: CAC defines a
set of action taken by the network during call set-up or re-negotiating phase
in order to determine whether a VCC or VPC request can be granted or not.
Apparently routing is part of CAC actions. One important class of applications
of ATM networks is interactive programs such as videoconferencing. In order to
provide QoS requirements for all connections, CAC is an essential mechanism.
Traffic Shaping (TS) [45, 46]: A key element of the traffic contract from the user
perspective is that the sequence of cells can be sent to the network after a desired
modification of the traffic characteristics and is still compliant with the traffic
parameters in the contract. In other words, the user equipment can process the
source cell stream such that the resultant output to the network is conforming
to the traffic contract. The most popuiar traffic shaping algorithms as proposed
in literatures include Peak Cell Rate Reduction (PCRR) and Buffering. PCRR
can be achieved by operating the sending terminal at a peak rate less than
that in the traffic contract, reducing the possibility of conformance violation.
Buffering operates in combination with the leaky bucket algorithm to ensure
that cells will not violate the traffic parameters of the contract by buffering
cells until leaky bucket would admit them.
Priority Control [47]: By using Cell Loss Priority bit, the user may generate
traffic flows with different priorities. An ATM switch may selectively discard


BIOGRAPHICAL SKETCH
Wen-Yen Fn was born on March 4, 1965, in Taiwan. He graduated from
Taipei Municipal Chien-Kuo Senior High School in 1983. He received his Bachelors
degree in electrical engineering from Sun Yat-Sen University, Kao-Hsiung, Taiwan
in 1987 and his M.S. degree in electrical engineering from Chiao-Tung University,
Hsinchu, Taiwan in 1989. From 1989 to 1991, he served as a second lieutenant at the
communication group of the military police headquarter, Taipei, where he developed
a communication equipment supply system. He has been a graduate student at the
University of Florida since August 1991. During this time he has been a teaching
assistant, research assistant, and the system administrator of the Laboratory for
Information Systems and Telecommunications. He is expecting to receive his doctoral
degree in electrical enginnering in May 1998 and is currently a student member of
the Institute of Electrical and Electronics Engineers (IEEE).
121


CHAPTER 6
PID FLOW CONTROL SCHEME FOR VBR TRAFFIC
In communication networks with large delay-bandwidth product, congestion
could happen over shorter time scales than those at which end-to-end protocols such
as congestion control schemes typically operate. In such cases, the congestion can
dissipate rapidly before congestion feedback information returns to the source. Net
work designers, therefore, face a challenge. The bursty and cyclic nature of Variable
Bit Rate (VBR) traffic creates another issue for transmission in ATM networks. To
reach the dual goals of keeping cell loss rate low and network utilization high, we pro
pose an adaptive rate-based flow control scheme for real-time VBR traffic in ATM
networks.
The goal of the scheme is to minimize the impact of traffic overload in order
to limit the cell loss rate to an acceptable range and also increase the network uti
lization. The proposed flow control scheme is based on predicting the evolution of
buffer occupancy over time using a Proportional-plus-Integral-plus-Derivative (PID)
controller and a linear predictor to adaptively update the optimum data emission
rate at the transmitter. The adaptive policy attempts to keep the buffer occupancy
for each virtual channel at a steady level and the simulation results show that the
proposed scheme works effectively against network congestion. Along with the design
of the new flow control scheme, we also develop a hierarchically structured testbed
to measure network performance and explore various flow control schemes in ATM
networks with diverse classes of incoming traffic.
73


55
Traffic model
rate-based 1
A%
rate-based 11
A%
Poisson
0.0934
-
0.0973
-
SSP, H=0.7
0.0869
7.2%
0.0914
6.5%
SSP, H=0.9
0.0825
13.2%
0.0868
12.1%
Table 4.1: Performance comparison of Poisson process and SSP traffic models
0.14
0.12
0.1
- 0.08
Q-
§0.06
0.04
0.02
0
0 2 4 6 8 10
simulation time (sec)
Figure 4.3: Performance comparison of different flow control schemes
model for Self-similar traffic with H=0.7 renders 7.2 % and 6.5% of over-estimations
of the goodput for the BECN scheme without and with slow-start, respectively. While
using a Poisson process model for Self-similar traffic with H=0.9 results in 13.2 % and
12.1% of over-estimations of the goodput for the BECN scheme without and with
slow-start, respectively. These results demonstrate the failure of Poisson processes as
traffic models for traffic exhibiting self-similarity.
4.4 Classical Control Theory and Congestion Control
For a system as shown in Figure 4.6, classical control theory is Tisually valuable
as a source of insight in the design of feedback controllers. In a feedback flow control


ITZJ
Session Name:|LIST test run
Description:
Sdr: Create New Session
This is a multicast test run in the Lab. for Information Systems and
Telecommunications.
it
y
URL: www.list.ufl.edu
Test URL
Encryption
Type of Session:
Test
Scope Mechanism:
TTL Scope
v Admin Scope
Scope
Site
v Region
v World
15
Media:
\f\ fi audio
Protocol
RTP
Format
PCM
4
video
RTP
H.261
X|
whiteboard
XI
text
Address
Port
¡224.2.142.130¡29710
224.2.154.26
57770
0 Session will be active:
Once fnom:|Tue11 Nov | at ¡16:00 J for | 2 hours J
; from: j
; from: I
at
' for
at.
: for
Repeat for:
Person to contact about this session:
Ol|Wen-Yen list.ufl.edu>
S|wen-Yen 352-392-2584
Create
Show Daily Listings
Help
Dismiss
Figure 7.5: Multicast session with audio/video streams


CHAPTER 8
CONCLUSION
8.1 ATM Networks and Self-similarity
Asynchronous Transfer Mode (ATM) has been widely considered to be one
of the most promising solutions for the next generation high speed and multimedia
networks. The anticipation is mainly due to the scalable building blocks it provides
to serve as a stable foundation for evolving public and private networks. ATM is
designed to handle many kinds of information including voice, data, image and video
in an integrated manner, and is expected to support applications with diverse traffic
characteristics and quality of service (QoS) requirements
The discovery of self-similar behaviors in Local Area Networks (LANs) and
Wide Area Networks (WANs) is one the most remarkable milestones in the study of
communication networks of late. To facilitate the design and optimal performance
of functional ATM networks, two key issues are addressed in this Ph.D. research,
namely an appropriate characterization and modeling of ATM traffic, and effective
congestion management schemes suitable for the self-similar or fractal-like traffic ex
pected on ATM networks. To study network dynamics and real traffic characteristics,
we conducted traffic measurements and analysis based on empirical traces collected
in the Laboratory for the Information Systems and Telecommunications (LIST). We
found that the traces displayed self-similar characteristics. In addition, the simula
tion results in our research confirmed that the traffic in LAN/WANs and the traffic
generated by VBR service in ATM networks are more appropriately modeled by pro
cesses based on self-similar instead of traditional Poisson-based processes. In other
106


14
Service Class
Class A
Class B
Class C
ClassD
Delay Sensitive
X
X
Constant Bit Rate
X
Variable Bit Rate
X
Connection-oriented
X
X
X
X
Connectionless
X
X
AAL
AAL1
AAL2
AAL3/4
AAL5
Example
DS1, DS3
MPEG Video
SMDS
TCP/IP
Table 2.2: Service Characteristics of AAL Layers
at the end of a frame which may comprise np to 1365 cells. It has been shown
that AAL5 is at least as effective as AAL3/4 in detecting misordered or lost
cells.
Table 2.2 concludes the characteristics of all AAL service classes and their
corresponding layers.
2.2 Quality of Service
ATM networks are expected to be one of the most dominant information
infrastructures in the near future. However, many issues still need to be resolved
before they can meet the glowing expectation. One major challenge, for example,
is to design an efficient CAC and congestion control mechanism suitable for diverse
traffic characteristics in the networks. To facilitate efficient and functional networks,
comprehensive of the characteristics and QoS requirements of the traffic to be carried
is of primary importance.
The primary objective of ATM is to provide the cell transfer in a connec
tion with guaranteed performance. The performance parameters, or QoS classes, are
summarized in Table 2.3. A QoS class with a connection can have either specified
or unspecified performance parameters. A Specified QoS class specifies a set of per
formance parameters and the target value for the corresponding parameter. In the


69
(H, e) = (0.7,10~8)
(H, e) = (0.8,10-8)
(H, e) = (0.9,10-8)
genuine H
estimated CLP
genuine H
estimated CLP
genuine H
estimated CLP
0.66
3.2*10~9
0.76
1.7*10~9
0.86
4.7*10~9
0.67
5.2*10-9
0.77
1.6*10-9
0.87
3.9*10~9
0.68
4.7*10-9
0.78
3.2*10-9
0.88
6.7*10~9
0.69
6.1*10-9
0.79
4.1*10-9
0.89
6.3*10-9
0.70
7.4*10-9
0.80
6.4*10-9
0.90
8.5*10~9
0.71
7.7*10-9
0.81
6.8*10~9
0.91
7.7*10~9
0.72
8.4*10-9
0.82
8.5*10-9
0.92
9.4*10-9
0.73
8.7*10-9
0.83
9.6*10~9
0.93
8.7*10-9
0.74
1.2*10-8
0.84
1.3*10-8
0.94
9.7*10~9
0.75
1.3*10-8
0.85
1.6*108
0.95
1.2*108
Table 5.1: Robustness of the proposed CAC scheme
Assume a finite buffer size, 500 cells, and the parameter (u, v) = (1000,100).
Figure 5.4 shows the number of admitted connections given a certain cell loss prob
ability. Note that the number of admitted connections increases significantly as the
cell loss probability increases when we have a larger Hurst parameter. When H is
equal to 0.5, the number of admitted connections remain nearly unchanged as the
cell loss probability increases.
It is of particular interest to investigate the robustness of the proposed CAC,
supposed that the submitted Hurst parameter is different from the genuine value.
From Table 5.1 and Figure 5.2, we found that the requested CLP can be ensured as
long as the genuine Hurst parameter is at least 0.04 less than the submitted value.
This robustness is related to the fact that the CLP is derived from (5.26), which is
an upper bound of the accurate value.
If the peak rate allocation scheme is used as the CAC, then the number of
admitted connections is equal to the ratio of the link capacity to the PCR. Accord
ingly, the number of admitted connections in this case is 110,000/2000 = 55. From
Figure 5.4, it is clearly that the proposed CAC algorithm achieves more admitted
connections. Let us define the statistical multiplexing gain (SMG) as the ratio of the
number of admitted connections by using the peak rate allocation to that by using


29
a
1
P
Figure 3.1: 2-state Markov chain
3.1.3 Markov-Modulated Process Models
Markov-modulated process models constitute an extremely important class of
traffic models. The idea is to introduce an explicit notion of state into the description
of a traffic stream an auxiliary Markov process is evolving in time and its current
state controls (modulates) the probability law of the traffic mechanism..
Now assume that while X is in state i, the probability distribution func
tion (pdf) of arrivals is completely determined by i, and this holds for every? G
{0,1,2,..., m}. Note that when X undergoes a transition to state j {0,1, 2,..., m},
then a pdf for arrivals takes effect for the duration of state j, and so on. Thus the
pdf for arrivals is modulated by the state of X .
The most commonly used Markov-modulated process is Markov-Modulated
Poisson Process (MMPP) which combines the simplicity of the modulating (Markov)
process with that of the modulated (Poisson) process. In this case, the modulation
mechanism simply stipulates that in state i of X, arrivals occur according to a Pois
son process at rate AAs the state changes, so does the rate. As a simple example,
consider a two-state MMPP model, where one state is an on state with an associ
ated positive Poisson rate, and the other is an off state with associated rate zero
(such models are also known as interrupted Poisson). These models have been widely


40
<£>(!)) = 1 a\D ... apDp (3.31)
and
0(D) = b0 biD ... bqD\ (3.32)
where D is a delay operator, i.e., DkXn = Xn-k- It follows that (3.30) can be rewritten
as
<$>{D)Xn = Q(D)wn, (3.33)
or
Xn = $~1(D)Q(D)wn. (3.34)
We now turn to fractional ARIMA time series. Let A be the difference opera
tor, defined by AXn = Xn ATn_i = (1 D)Xn. A fractional ARIMA(p,d,q) process
is defined as a stochastic process X = (Xk : k = 0,1,2,...) represented by
${D)AdXn = Q(D)wk. (3.35)
Note that the parameter d in Ad is allowed to take fractional values, either positive
or negative. Logically the fractional difference operator follows as
Ad = (1-D)d = Y.C(d,k){-D)k, (3.36)
k=1
where
k b n 1
c(d,k) = n
3=1
T(k d)
(3.37)
j r(-d)r(fc +1)'
For example, if p=q=0 and d is a non-negative integer, then AdXn = wn describes
a model where Xn< differenced d times, yields a sequence of i.i.d. random variables
wn. As a consequence, ARIMA(0,1,0) characterizes a random sequence generated by
random walk. Of particular notice is it has been shown in [34] that for d e (0,1/2)


102
Protocol
p (packet/sec)
a (packet/sec)
H
MBONE
29.22
52.10
0.50
TCP
167.78
2457.4
0.78
UDP
24.20
807.81
0.87
ALL
333.82
4239.84
0.81
Table 7.1: Arrival rate statistics
0,
0)
o
c
CO
CO
>
"O
CD
N
1
E
O) _
O -2
-0.5
-1
-1.5
-2.5
0
3T
IT
p:
*:MBONE
O: TCP
+: UDP
x: ALL
+
i X + +
Q n X v
r H=0.9
+ +
X
+
0 x ^ 1
u o X x H=0B
o
o >
H=0.7 ^
H=0.6
H=0.5
*
0.5
1
1.5
log10(Aggregation level (m))
Figure 7.10: Variance -time plot
Long Range Dependence (LRD). In addition, the estimated Hurst parameter for the
overall traffic is 0.81, which implies LRD as well.
Figure 7.11, Figure 7.12, and Figure 7.13 show the Quantile-Quantile plot
(QQ-plot) for MBone, TCP, and UDP traffic, respectively. The purpose of the QQ-
plots here is to determine whether the traces come from a Poisson distribution re
gardless of parameter values. If the trace does come from a Poisson distribution, the
plot will be linear. Otherwise, a curvature will be introduced in the plot. It is evident
that both the TCP and UDP traces deviate radically from a Poisson distribution,
while the MBone trace is matched by a Poisson distribution quite well. Figure 7.14


48
cells with low priority if necessary to protect the QoS of cells with high priority
as much as possible.
Usage Parameter Control (UPC) [48] is defined as the set of actions taken by
the network to monitor and control traffic in terms of traffic offered and fairness
of ATM connections. The main purpose is to protect network resource from
malicious or unintentional misbehaviors, which can affect the QoS of other al
ready established connections, by detecting violations of negotiated parameters
and taking actions accordingly.
Feedback Control [6, 13, 49, 8] is defined as the set of actions taken by the
network and by the users to regulate the traffic submitted on ATM connections
according to the state of switch such as the dynamic value of Cell Loss Rate,
Cell Transfer Delay or buffer occupancy.
One way to classify congestion control schemes is by the layer of ISO/OSI
reference model at which the scheme operates [50]. For example, there are data link,
routing/networking, and transport layer congestion control schemes. Typically, a
combination of such schemes is used. For sporadic congestion, CAC is used to route
according to load level of the links and to reject new connections if all paths are highly
loaded. For congestion lasting less than the duration of connection, an end-to-end
control scheme can be used. For example, during connection setup, the sustained
and peak rate may be negotiated. A leaky bucket algorithm may be used later by the
source or the network to ensure that the input meets the negotiated parameters. Such
traffic shaping algorithms are open loop in the sense that the parameter cannot be
changed dynamically if congestion is detected after congestion. On the contrary, in a
closed loop scheme, sources are informed dynamically about the congestion state or
resource usage of the network and are asked to adjusted their input rate accordingly.
In general, this information is carried by acknowledgment or RM cells via hop-by-hop


25
high speed switching fabric to accommodate simultaneous arrivals from different in
put ports. The switching fabric and output buffer must effectively operate at a much
higher speed than that of each interface port. The implementation of the high-speed
bus and buffer memory could be complex if the required memory access speed is very
high, thus limiting the capability of the switch to support very high speed interface
ports.


8 CONCLUSION
106
8.1 ATM Networks and Self-similarity 106
8.2 Preventive Flow Control 107
8.3 Feedback Flow Control 107
8.4 Future Work 108
APPENDICES
A SCRIPTS FOR GENERATING SELF-SIMILAR TRAFFIC 110
A.l Random Mid-point Displacement Algorithm 110
A.2 Power Spectrum Density Approximation Ill
B SCRIPTS FOR ESTIMATING HURST PARAMETER 113
B.l Variance Time Plot Algorithm 113
B.2 Rescaled Adjustment Algorithm 114
REFERENCES 116
BIOGRAPHICAL SKETCH 121


68
Figure 5.1: Cell loss probability vs. buffer size
cell loss probability decreases significantly as the buffer size increases. However, when
H=0.8 or H=0.9, increasing the buffer size merely renders a margin of improvement
in the cell loss probability. This is contradictory to the results obtained from generic
queueing models which do not take the Hurst parameter into account.
To study the sensitivity of the proposed CAC scheme to the changes in buffer
size, we calculate the number of admitted connections for different buffer size. The
parameters for the traffic source are (u, v, e) = (1000,100,10~8), PCR=2000 cells/sec,
and the output port is a DS-3 link at the speed of 110,000 cells/sec. The number
of admitted connections vs. buffer size is plotted in Figure 5.3 where the number of
admitted connections decreases as the Hurst parameter increases. This is because the
greater the Hurst parameter, the burstier the traffic is, and consequently the larger
allocated bandwidth is required.


109
on the fly. It poses a challenge especially for estimating the Hurst parameter
which requires longer time interval to obtain more accurate estimation.
Feedback flow control scheme:
It is well-known that an Integral controller is usually used in conjuction with
a Proportional controller to reduce the steady-state error, and a Derivative
controller is often added to a PI controller to increase the system stability. It
is, therefore, desirable to study and verify the system behaviors of the proposed
feedback flow control scheme in terms of relative stability and steady-state
accuracy corresponding to different parameters. To improve the performance
of the flow control scheme, other sophisticated control designs such as Model
based Predictive Control (MPC), and predictors based on AR model or fuzzy
logic can be used at the expense of complexity. Of particular note is the fact
that both effectiveness and simplicity are key requirements of a flow control
scheme, so the trade-offs are worth further investigation.


57
b*
Controller
Uncertainty
Network
b
+
A
\
u
Feedback
Figure 4.6: Generic Feedback Control System
system, this figure is the type of system used to ensure that the network, the con
trolled object characterized by input u, internal states x, output y, and unavoidable
uncertainty should perform in a desired manner.
Several important insights from classical control theory are as follows.
1. A feedback controller must combine dual function of estimation and of control.
The estimation reduces uncertainty about current state whose current values
are needed in order to determine current control action.
2. Feedback is the means by which control is achieved in the presence of uncer
tainty. Therefore, any feedback system should benefit from a design approach
which take explicit account of uncertainty.
3. Two different types of uncertainty can affect controlled objects: uncertainty
about current values of the internal states; and uncertainty about future val
ues resulting from uncertainty about controlled object dynamics governing the
evolution of future state.
The concept drawn from the classical control theory will serve as a basis for
the proposed flow control schemes in Chapter 6.99


18
model provides an accurate approximation of the bit rate of a single video source
without scene changes, but is not quite useful for queueing analysis.
Figure 2.3: TV Comercial
2.3.3 ABR Traffic
Available bit-rate (ABR) traffic is usually referred to data applications, i.e.,
any application that is not voice, audio, video, or still image. Despite the fact that
data networks have been operational for a number of decades, traffic characteristics
of data sources are not well understood. The main difficulty comes from the fact
that there does not exist any typical data connection. The problem of characterizing
the source characteristics of ABR traffic is further complicated by the difficulty of
predicting in advance the traffic characteristics of a connection, even if the particular
application type is known. For example, in client-server computing, the amount of
data exchanged and the source behavior may differ significantly from one application
to another. Furthermore, data connections are not generally established between
two users, but between a group of users, as in the case of local area network (LAN)
interconnection.


118
[29] R. G. Addie, M. Zukerman, and T. Neame. Fractal traffic: Measurements, mod
eling and performance evaluation. In Proceedings of the Conference on Computer
Communications (IEEE Infocom), Boston, Massachusetts, April 1995.
[30] B. B. Mandelbrot. Self-similar error clusters in communication systems and the
concept of conditional stationarity. IEEE Trans. Communications Technology
COM-13, 13:71-90, 1965.
[31] B. B. Mandelbrot and J. W. Van Ness. Fractional Brownian motions, fractional
noises and applications. SIAM Review, 10:422-437, 1968.
[32] S. Samorodnitsky and M. S. Taqqu. Stable Non-Gaussian Random Processes.
Chapman and Hall, second edition, 1994.
[33] H. E. Hurst. Methods of using long term storage in reservoris. Proc. of the
institution of Civil Engineers, part 1:519-577, 1955.
[34] D. R. Cox. Long-range dependence: A review. Statistics: An Appraisal, pages
55-74, 1984.
[35] B. Mendelbrot. Long-run linearity, locally gaussian process, h-spectra and infi
nite variances. International Economical Review, pages 82-113, 1969.
[36] C. Huang, M. Devetsikiotis, I. Lambadaris, and A. R. Kaye. Self-similar model
ing of variable bit rate compresses video: A unified approach. ACM SIGCOMM,
1995.
[37] C. W. J. Granger and R. Joyeux. An introduction to long-memory time series
models and fractional differencing. J. Times series Anal, 1:15-29, 1980.
[38] V. Paxson. Fast approximation of self-similar network traffic. Technical Report
LBL-36750, Lawrence Berkeley Laboratory and EECS Division, University of
California, Berkeley, California, April 1995.
[39] W. Leland, M. Taqqu, W. Willinger, and D. Wilson. Self-similarity in high
speed packet traffic: Analysis and modeling of Ethernet traffic measurements.
Statistical Science, 1994.
[40] G. M. Bernstein and D. H. Nguyen. Blocking reduction in fast reservation pro
tocols. In Proceedings of the Conference on Computer Communications (IEEE
Infocom), Toronto, Canada, June 1994.
[41] S. L. Bermejo and G. H. Petit. Bandwidth resource dimensioning in ATM net
works: A theoretical approach and some study cases. In 13th International
Teletraffic Congress, volume 14, pages 929-934, Copenhagen, June 1991.
[42] K. Kang, Y. Yoon, and C. Kim. A CAC scheme for heterogeneous traffic in
ATM networks to support multiple QoS requirements. In Conference Record of
the International Conference on Communications (ICC), volume 1, pages 422 -
426, 1995.
[43] J. J. Harms and Q. Hong. A call admission scheme for ATM networks based
on refinement of traffic measures. In Conference Record of the International
Conference on Communications (ICC), volume 1, pages 191 195, 1995.


93
Sender
Receiver #1 Receiver #3
Receiver #2
Figure 7.2: Reservation Merge of RSVP
ATM allowed in a single point to a multipoint Virtual Channel (VC). At a minimum,
the Internet model allows both best effort and QoS service for the same multicast
session, and in general allows each receiver to have a different QoS. Second, RSVP
supports dynamic QoS in reservations, meaning that users may change their QoS at
any time.
7.1.2 Real Time Transport Protocol
Due to its unpredictable delay, TCP, contrasting to UDP, is unsuitable for
applications with real-time characteristics. To use UDP as a transport protocol for
real-time traffic, however, some functionality has to be added. The functionality
that is needed for many real-time applications is incorporated into the Real-time
Transport Protocol (RTP) [68, 69, 70].
The RTP is a protocol providing support for applications with real-time prop
erties, including timing reconstruction, loss detection as well as security. RTP helps
ensure reliable, real-time delivery by adding fields containing timestamp and sequence


80
(a)
r[n]
&
Compensator
Plant
y[n]

(b)
b*
0
Controller
Network
Figure 6.3: Generic tracking system and its application in networks
the reference r[n] and output y[n] as a trigger generates a dynamical signal entering
the plant. The compensator must be well designed so y[n] can track r[n] promptly.
One of the most well-known compensators is probably the so-called Proportional-
plus-Integral-plus-Derivative (PID) controller. Figure 6.3(b) shows the scenario that
applies the tracking system to networks. In this chapter, we exploit this system
theoretic methodology in our proposed flow control scheme.
Figure 6.4: PID flow controller


91
Traffic measurements and analysis based on the traces collected in the Laboratory
for Information Systems and Telecommunications (LIST) is presented in Session 2.
7.1 Protocols for Integrated Services
7.1.1 Resource Reservation Protocol
RSVP is a resource reservation setup protocol designed for an integrated ser
vices Internet. The RSVP mechanisms provide a facility to create and maintain a
distributed and dynamic reservation state over a multicast or unicast transmission
path. RSVP also provides many other important features for signalling in the In
ternet including receiver oriented reservations, heterogeneous reservations within a
multicast session, and multiple reservation styles. Several of these features directly
impact ATM signalling and IP over ATM signalling, which will be described in more
detail later in this session.
To make a resource reservation at a host with RSVP server installed, the
RSVP server needs to inform two local modules first: Policy Control and Admission
Control as shown in Figure 7.1 [67]. The policy control determines whether the
user has the privilege for reservation, and the admission control decides whether
resources are available for the requested QoS. An error notification will be returned
to the application process if any request fails. The request will be granted only if
both checks get successful. The RSVP server will then continue to set parameters
in a Packet Scheduler and Packet Classifier to acquire the desired QoS. The packet
classifier assigns a QoS class to each packet according to the parameters, and the
packet scheduler maintains the promised QoS for the each stream during packet
transmission.
When the RSVP session is a multicast, RSVP defines merging of reservations
on network links where a single flow is destined to multiple receivers. Just as IP
multicast routing avoids the need to transmit multicast data more than once on a


8
Figure 2.1: B-ISDN Protocol reference model
and control functions for each network and its endpoints. The primary layers of the B-
ISDN protocol reference model are the Physical Layer that provides for transmission
of ATM cells over a physical medium that connects two ATM devices, the ATM Layer
where cell segmentation and reassembly occurs, and ATM Adaptation Layer (AAL)
that provides support for higher layer services such as circuit emulation, frame relay
and SMDS.
2.1.2 Physical Layer
The Physical Layer of the ATM stack is divided into two sublayers: the Trans
mission Convergence (TC) sublayer and the Physical Medium (PM) sublayer. The
TC sublayer is responsible for such tasks as line coding and bit timing, the generation
and verification of the header error control byte, cell mapping and cell delineation
and the PM sublayer provides bit-transmission capabilities on any particular medium
including optical fiber, coaxial cable, Unshielded Twist Pair (UTP), and radio links.
Users have quiet a range of Physical Layers to choose from. The ATM Forum
[20] has issued recommendations for 100 Mpbs and 140 Mbps transmission over multi-
mode optical fiber, a 155 Mbps Synchronous Optical Network (SONET) interface,
a 45 Mbps T3 interface, and is in the process of drawing up recommendations for


45
3.4.5 Approximation of Power Spectrum
The method described in [38] is based on the Fast Fourier Transform. The
mathematics of this method are beyond the scope of this research, but the strategy
behind the method is to generate a sequence of complex numbers corresponding to the
power-spectrum of fractional Gaussian noise. The inverse Discrete Fourier Transform
is then used to obtain the time-domain counterpart of this power-spectrum. Because
autocorrelation and power-spectrum form a Fourier pair, the resulting process is guar
anteed to have the autocorrelation structure (and, hence, self-similarity) of fractional
Gaussian noise. This method has the property that it is fast (generating 262,144
points takes 80 seconds on a SPARCstation IPX) and that it does not suffer from
the biases associated with the Random Midpoint Displacement algorithm. Because
it is based on the FFT, this method is referred to by its authors as the FFT method
of generating self-similar sample paths. A Matlab script based on this method is
included in Appendix B.2.


84
remaining 20% of link speed, which is about 9 Mbps or 21,227 cells/sec. The trace
of the video used in the simulation is shown in Figure 6.1. The frame rate of the
video, 1/tf, is 30 frames/sec. In addition, the PCR and SCR is 7,700 cells/sec and
l,700cells/sec, respectively. The normalized network utilization rate p^ is obtained
by estimating the ratio of average VBR bandwidth to the overall bandwidth available
to VBR. To simplify our simulation efforts and put the emphasis on the impact of
K/ and K^, the value of l in Equation 6.4 is set to be 5.
Figure 6.7 and Figure 6.8 show the simulation results where the proposed
flow control scheme is disabled. Figure 6.7(a) and Figure 6.7(b) show the frame size
transmitted at CPE-T1 and that received at CPE-D1, respectively. The CLR (Cell
Loss Rate) defined by the ratio of NT, the total number of cells transmitted, to NLl
the total number of cell lost, can be obtain from Figure 6.8 (a) and Figure 6.8 (b).
The CLR in this case is 0.046 (Afy = 19285, NL 887). The CLR is clearly less than
5% without using the rate controller, but p^, the normalized network utilization rate,
is as low as 30% which means the bandwidth available to VBR traffic is unused 70%
of the time.
Figure 6.9 and Figure 6.10 show the results for the same statistics as those
of Figure 6.7 and Figure 6.8 when the proposed flow control scheme is used with
(K/, Kd) = (0.2, 0.05). Notice that, with more cells transmitted (and better video
quality) than the case of no feedback, the CLR increases to 0.072 (Afy = 20800, NL =
1488), but pN is improved substantially to 43%. The results under different values of
(Kj. Kq) are listed in Table 6.1. Both CLR and p^ increase as the values of Kj,Kd
increase.
Another opportunity to take advantage of the proposed flow control scheme
and make the most use of available bandwidth is to aggregate more VBR streams. Let
Nvbr denote the number of VCs for video at each CPE. The CLR and pN corresponding
to different value of N^r are shown in Table 6.2. As Nbr increases, more bursty video


104
Figure 7.12: QQ-plot for TCP traffic
Figure 7.13: QQ-plot for UDP traffic


43
Figure 3.4: Fractional Brownian motion and fractional Gaussian noise


17
situation and still remain within the range of good quality, but for some mission
critical data service, even a single bit of data lost could result in the redundancy of
the entire data received.
2.3.1 CBR Traffic
CBR traffic, generated by existing circuit switched systems as well as pack-
etized voice and video sources using fixed rated coders, will be a dominant traffic
in evolving broadband networks. Such traffic types are characterized by stringent
delay requirements. Because of the periodic nature of the traffic, if a packet from a
given source experiences a long delay (or is blocked due to the buffer overflow), then
successive cells from the same source would experiences the same delay (or will be
blocked) until the superposed arrival pattern undergoes a change.
2.3.2 VBR Traffic
The traffic generated by a typical source, in general, either alternates between
active and silent periods and/or has a varying bit rate generated continuously. Fur
thermore, the peak-to-average bit rate of a VBR source is often much greater than
one. Presenting VBR traffic to the network as CBR traffic by means of buffer smooth
ing has the drawback of underutilization of network resources and QoS degradation.
Although doing so would simplify the network management task, it is more natural
to provide VBR service to VBR sources to achieve higher resource utilization.
The characteristics of the video signal depend primarily on two factors, the
nature of the video scene, and the type of VBR coding technique employed. Figure 2.3
and Figure 2.4 show the frame size over elapsed time of two typical MPEG 11 videos
generated from recorded TV news and commercial. Several models were introduced
to characterize video traffic with and without scene changes. For instance, a first
order AR model was used to approximate the output rate of video sources. The


52
4.3 Impropriety of Poisson-Based Traffic Models
This session presents the simulation efforts to justify the performance of the
flow control schemes introduced in the previous section by using Poisson process
traffic models and self-similar process (SSP) traffic models. The RMD algorithm is
used to generate a self-similar sample path. Since fGn has zero mean, it can not be
directly used to represent a TCP packet size. However, since a linear transformation
does not affect Hurst parameter, it may be used to make the sample paths eligible
for a packet size. Note that due to the re-transmission of erroneous packets in TCP,
any single cell loss of a packet in ATM networks will result in the discard of any
other successfully transmitted cells of this packet. The effective throughput taking
into account the cell discard is so-called goodput.
4.3.1 Network Configuration
Figure 4.1 shows the network configuration of the simulation scenario where
two CPS-100 ATM switches are connected to each other by a trunk link at a speed of
DS-3 (44.5 Mbps). Eight CPEs are attached to each of the two switches. We assume
a one-to-one correspondence among each pair of transmitting and receiving CPEs.
In other words, a virtual path or channel connection is established between each
CPE attached to CPS 100-A and the corresponding CPE attached to CPS 100-B.
The RMD algorithm is iised to generate synthesized self-similar traffic with a target
Hurst parameter at each CPE attached to CPS 100-A. The self-similar data flow is
then destined for the CPS 100-B through the trunk link. Note that the CPEs attached
to CPS100-B send only RM cells with congestion information through feedback. As
active traffic sources aggregate on the outgoing trunk port of CPS100-A, the egress
buffer will build up eventually if there are too many arriving cells in a short period.
The egress buffer is, therefore, a performance bottleneck. Due to balanced traffic
and symmetric network topology, the same value of effective throughput, namely,


78
Video sender Video receiver
Figure 6.2: Generic Video Transmission System with Feedback
constraints imposed by the encoder and decoder buffers, the leaky bucket con
trol, traffic smoothing, and rate control. The encoder rate control is separated
into a sustainable-rate control and a unidirectional instantaneous-rate control.
Lee and et al. [62] suggested a modified explicit rate indication for conges
tion avoidance (MERICA) scheme for both VBR and CBR traffics in ATM
networks. This modified scheme belongs to the class of closed-loop rate-based
congestion control scheme.
Adaptive credit-based flow control: Goyal and et al [63] presented a receiver-
oriented adaptive credit-based flow control scheme to minimize the buffer re
quirement in the network while guaranteeing that cells of VBR encoded video
flows will not be lost. With the proposed bandwidth estimation techniques
which exploit the structure of the video traffic, it also minimizes the end-to-end
delay and jitter for VBR encoded video streams.
Data recovery: Lost data may be recovered using temporal and spatial interpo
lation or forward error recovery methods. Hu and Lin [64] suggested a forward
error control scheme for MPEG video to recover lost cells in ATM networks.
The original MPEG streams are split into n substreams and the substreams are


60
call setup procedure has to be carried out before a user starts transmitting over an
ATM network. The main objective of this procedure is to establish a path between
the sender and the receiver. In addition, this procedure also allocates resources in
every switch along the path to accepted connections. Generally, CAC deals with the
question whether or not a switch can accept a new connection. CAC may be classified
into two categories, namely, statistical allocation and non-statistical allocation. Non-
statistical allocation is also referred to as so-called peak rate allocation.
The effectiveness of CAC mainly depends on appropriate traffic modeling and
characterization. In this chapter, we investigate the behaviors of an ATM queueing
system with a self-similar arrival process. The service process is deterministic and the
arrival process is modeled by a fractional Brownian motion (fBm) process. The main
goal of this chapter is to design a new effective CAC suited for services exhibiting
self-similarity.
The remaining of this chapter is organized as follows. Section 2 and Section
3 describe peak rate allocation and statistical allocation scheme, respectively. In
Section 4, we develop a queueing model to evaluate the cell loss probability for a
system with fractional Brownian motion arrival processes. In Section 5, we propose
a CAC based on the results obtained from Section 4. Numerical results are presented
in Section 6, and conclusion is given in Section 7.
5.2 Peak Rate Allocation
Peak rate allocation (PRA), also called non-statistical allocation, is the most
widely used CAC due to its simplicity. When a user requests a connection, the
decision to accept or reject the request depends on whether its peak cell rate (PCR)
is greater or less than the available rate The major drawback of PRA is that the
network resources will be well under-utilized if the ratio of PCR to MCR (mean cell
rate) of the connection is much larger than one. Using PRA for VBR services is one
of the best examples of this situation. Given a PCR/MCR equal to R, the resource


119
[44] C. J. Chang and R. G. Cheng. Traffic control in an ATM network using fuzzy set
theory. In Proceedings of the Conference on Computer Communications (IEEE
Infocom), Toronto, Canada, June 1994.
[45] S. Manthorpe and X. Garcia. Shaper dimensioning for MAN to ATM based
WAN. In Interop94 Engineers Conference, Las Vegas, California, May 1994.
[46] G. de Veciana and J. Walrand. Traffic shaping for ATM networks. Draft, Depart
ment of Electrical Engineering and Computer Sciences, University of California,
Berkeley, California, December 1992.
[47] P. Skelly, G. Pacifici, and M. Schwartz. A cell and burst level control framework
for integrated video and image traffic. In Proceedings of the Conference on
Computer Communications (IEEE Infocom), Toronto, Canada, June 1994.
[48] A. R. Reibman and A. W. Berger. Traffic descriptors for VBR video teleconfer
encing over ATM networks. IEEE/ACM Transactions on Networking, 3(3):320-
328, June 1995.
[49] A. Atai and J. Hui. Rate-based feedback traffic controller for ATM networks. In
Conference Record of the International Conference on Communications (ICC),
volume 3, pages 1605-1615, New Orleans, Louisiana, May 1994. IEEE.
[50] M. Decina, T. Toniatti, P. Vaccari, and L. Verri. Bandwidth assignment and
virtual call blockings in ATM networks. In Proceedings of the Conference on
Computer Communications (IEEE Infocom), pages 881-888, San Francisco, Cal
ifornia, June 1990. IEEE.
[51] H. Esaki, K. Iwamura, and T. Kodama. A simple and effective admission control
method for an ATM network. In Proceedings of the IEEE Conference on Global
Communications (GLOBECOM), pages 28-33, San Diego, California, December
1990. IEEE.
[52] P. Boyer. A congestion control for the ATM. In International Teletraffic
Congress: Seventh Specialist Seminar, page 4.3, Morristown, New Jersey, Octo
ber 1990. ITC.
[53] H. T. Kung, R. Morris, T. Charuhas, and D. Lin. Use of link-by-link flow control
in maximizing ATM network performance simulation results. In Appeared in
Proc. IEEE Hot Interconnects Symposium, 93, page 12 pages, August 1993.
[54] I. Norros. A storage model with self-similar input. Queueing Systems, pages
387-396, 1994.
[55] R. Guerin, H. Ahmadi, and M. Naghshineh. Equivalent capacity and its applica
tion to bandwidth allocation in high-speed networks. IEEE Journal on Selected
Areas in Communications, 9:968-981, 1991.
[56] J. Walrand. An Introduction to Queueing Netwroks. Prentice-Hall, 1988.
[57] V. E. Benes. General Stochastic Processes in the Theory of Queues. Addison-
Wesley, 1963.
[58] N. G. Duffield and N. OConnell. Large deviations and overflow probabilities for
the general single-server queue, with applications. Proc. Cambridge Phil. Soc..,
1995.


120
[59] L. G. Roberts. Can ABR service replace VBR service in ATM networks. In
COMPCON 95. Technologies for the Information Superhighway, Digest of Pa
pers., pages 346-350, 1995.
[60] P. Pancha and M. E. Zarki. MPEG coding for variable bit rate video transmis
sion. IEEE Communications Magazine, pages 54-66, may 1994.
[61] W. Ding. Joint encoder and channel rate control of VBR video over ATM
networks. In IEEE Transactions on Circuits and Systems for Video Technology,
pages 266-278, 1997.
[62] C. B. Lee, K. B. Ha, and R. H. Park. Computation of effective bandwidth
of aggregated VBR MPEG video traffic in ATM networks using the modified
equivalent capacity. In Conference Record of the International Conference on
Communications (ICC), volume 2, pages 627 631, 1996.
[63] P. Goyal, H. M. Vin, C. Shen, and P. J. Shenoy. A reliable, adaptive network
protocol for video transport. In Proceedings of the Conference on Computer
Communications (IEEE Infocom), volume 3, pages 1080 1090, 1996.
[64] L. C. Hu and K. J. Lin. A simulation study of real-time MPEG traffic using
forward error control scheme object-oriented real-time dependable systems. In
Proceedings of WORDS96, pages 232-240, 1996.
[65] M. R. Ismail, I. E. Lambadaris, M. Devetsikiotis, and A. R. Kaye. Modelling pri
oritized MPEG video using TES and a frame spreading strategy for transmission
in ATM networks. In Proceedings of the Conference on Computer Communica
tions (IEEE Infocom), pages 762-770, 1995.
[66] C. S. Li, Y. Ofek, and M. Yung, time-driven priority flow control for real-time
heterogeneous internetworking. In Proceedings of the Conference on Computer
Communications (IEEE Infocom), volume 3, pages 1080 1090, 1996.
[67] R. Braden, L. Zhang, and et al. Resource reservation protocol version 1 func
tional specification. Technical report, IETF, December 1997.
[68] R. Frederick H. Schulzrinne, S.Casner and V. Jacobson. Rtp: A transport pro
tocol for real-time applications. In RFC 1889, number RFC 1889. IETF.
[69] H. Schulzrinne. Rtp profile for audio and video conferences with minimal control.
In RFC 1890, number RFC 1890. IETF.
[70] D. Hoffman and G. Fernando. Rtp payload format for mpeg/mpeg2 video. In
RFC 2038, number RFC 2038. IETF.
[71] K. C. Almeroth and M. H. Ammar. Multicast group behavior in the internets
multicast backbone (mbone). In IEEE Communications Magazine, pages 124-
129, June 1997.
[72] M. R. Macedonia and D. P. Brutzman. Mbone provides audio and video across
the Internet. In Computer, pages 30-36, April 1994.
[73] S. E. Deering and D. R. Cheriton. Multicast routing in datagram internetworks
and extended LANs. In ACM Transactions on Computer Systems, pages 85-110,
May 1990.
[74] Network Research Group Lawrance Berkeley National Laboratory. Tcpdump
3.4.


100
Figure 7.8: LIST-NET Topology
performance. Over the past few years, this tool has been steadily improved by the
excellent contributions from the Internet community. With Tcpdump, one can print
out the packets on a network interface in various expressions according to protocols,
port numbers, packet length, and so on. The time resolution of Tcpdump is 1 /sec.
7.2.2 Traffic Measurements and Analysis
We monitor and collect the packet flows in the LIST_Net by running Tcpdump
in a Linux host latchman. The packets are collected first and then filtered out
according to the algorithm defined in a shell script. The traces represent both LAN
and WAN traffic in a sense that the WAN traffic is corresponding to those packets
destined to or originated from external networks, while the LAN traffic are those
originated from and destined to the LIST_Net. The original trace is further divided
into several groups according to the protocol with which each packet is associated.
The protocols of interest are IP Multicast (MBone), Transport Control Protocol
(TCP), and User Datagram Protocol (UDP).


35
Figure 3.3: Autocorrelation function and long-range dependence
r(r) = t2h~2{2H(2H 1) + /(r)], (3.15)
where /(r) a\T~2 + a2r-4 + . Since /(r) = 0 as r > oo, we have
r(r, H) = H{2H 1 )r2H~2. (3.16)
From (3.16), it is obvious that r(-r) is summable if H < 0.5, and non-summable,
otherwise.
Figure 3.3 depicts r(r) for several values of H. Of particular notice is r (r) decays
to 0 as t > oo much slower when H > 0.5 than when H < 0.5. As a result, r(r)
tends to 0 so slowly that ^o r(r) = i which leads to the LRD of X(t).


I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
Haniph A. Latchman, Chairman
Associate Professor of Electrical and
Computer Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
Leon W. Couch II
Professor of Electrical and Computer
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
red J
Professor of Electrical and Computer
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
ftindy ChdvT
Professor of Computer and Information
Science and Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
-
Richard Newman-Wolfe
Assistant Professor of Computer and
Information Science and Engineering


13
become a key protocol in ATM implementation requiring support for VBR
audio and video, for example, Motion Picture Expert Group (MPEG) video.
In theory, this AAL could be used to synchronize separate streams of data
without the need for elaborate timestamping in the data streams themselves.
Due to the similar timing requirement as AAL1, AAL2 can also correct single
bit errors in the payload and notifies lost cells or misordered cells.
AAL3/4 is the result of standards efforts for two AALs: AAL3 and AAL4,
which are both for transport of VBR streams without explicit timing informa
tion. AAL3 is connection-oriented and AAL4 is connectionless, although it is
unclear what meaning this distinction actually has in ATM, and usually these
two services are lumped together. As it turns out, the (then) CCITT merged
AALs 3 and 4 into one single 3/4 type whose procedures can be applied in a
connection-oriented or connectionless manner. Both AAL3 and AAL4 provide
assured and non-assured services. In the assured mode, the AAL guarantees
delivery of AAL Service Data Units (SDU) in order, and any lost SDUs are
retransmitted. In the non-assured mode, this function is performed by higher
layers. The main advantage AAL3/4 has is that it carries a length indication
in its first cell, and, hence, make easier some traffic management function such
as fast buffer reservation.
AAL5, also known as Simple and Efficient Adaptation Layer (SEAL), was
developed in response to a perception that AAL3/4 was inefficient, since only
44 octets of a cell payload carry actual data. AAL5 is a substantially lean
AAL compared with AAL3/4 at the expense of error recovery and built-in
retransmission. The trade-off provides a smaller bandwidth overhead, simpler
processing requirements, and reduced implementation complexity. In AAL5,
there is very little AAL overhead, with just one header and one 32-bit checksum


CHAPTER 2
BACKGROUND
2.1 Asynchronous Transfer Mode
The ATM standard is primarily defined by International Telegraph and Tele
phone Consultative Committee (CCITT), lately called International Telecommunica
tions Union-Telecommunication standardization sector (ITU-T). Some interim stan
dards for some aspects of ATM have been developed by a user and vendor group
known as ATM Forum in the absence of CCITT standards [16, 17, 18]. ATM as
an emerging, cell-based technology is expected to integrate the currently separate
networks used for voice, video, and data applications. The merger of these networks
has the potential to provide us significant benefits and convenience. ATM, originally
bonded to emerging Telecommunications technology standards, is an established way
to provide channels with arbitrary bandwidth within a multiplexing hierarchy consist
ing of a well defined set of fixed bandwidth channels. Eventually, a side-effect of this
provision of arbitrary-capacity channels, ATM will also be able to support channels
of variable bit rate, and hence, will be capable of achieving a statistical multiplexing
gain (SMG). This service is currently available in speeds ranging from 1.544 Mbps
(T1 or DS-1) to 155 Mbps (Optical Carrier Level 3 or OC-3), with rates expected to
be available down to 64 Kbps and up to 622 Mbps (OC-12) soon. Ultimately, ATM
is expected to scale up to the gigabit range.
As a promising candidate for networking in LANs as well as a replacement
for Time Division Multiplexing (TDM) transmission systems in WANs, researchers
routinely saw ATM as a scalable method for the provision of high-speed network
6


71
requested CLP
estimated CLP with PC AC
estimated CLP with EC
1(T3
6.2*10-4
3.5*10~2
1(T4
5.9*10~s
4.6*10-3
1(T5
3.7*10~6
7.1*10-4
1CT6
8.1*107
1.3* 105
10"7
7.4*10~8
9.4*10~5
icr8
1.7*109
5.4*10-6
109
1.4*10-1
7.9*10-6
Table 5.2: Guarantee of QoS
120
100
a
CD
o
o
T3
CD
E
a
<
40
10'
10z
10'
Buffer size
10"
10'
Figure 5.3: Number of admitted connections vs. buffer size


32
3.2.1 Definitions
A wide-sense stationary (WSS) random process X(t) is defined as exactly self
similar [1] if a small portion of X(t) can exactly reproduce a large portion of X(t) by
scaling properly. What is of more interest to network performance analysis, however,
is the so called statistical self-similar [1], A WSS random process X(t) is (statistically)
self-similar if each smaller portion shares statistical properties of larger portion of X(t)
after properly scaled. A more explicit definition of self-similar in [32] is defined below.
Definition 1 X(t) is self-similar with Hurst parameter H G (0,1], if
X (t)=\a\~H X (at), Va (3.4)
and
X(t + r) X(t)=X(r) X(0), V r (3.5)
where = denotes the equivalence of distribution.
By this definition, a self-similar process has stationary increments as indicated
in (3.5). In addition, this process is also distribution invariant under appropriate scal
ing in time and space in a sense that the self-similarity can be observed by comparing
small bursts in a small region with larger ones, with smaller ones superimposed, over
a larger region of the horizontal axis. This distribution invariant characteristic is
observed in an empirical trace show in Figure 3.2 as well. The variations of the defi
nition of self-similarity which lead to more important properties will be described as
follows.
Let X(t) be a WSS random process with mean /i, variance a1 and autocorre
lation function r(r). In particular, we asstime that r(r) is of the form
t2 2HS(t), as r > oo
(3.6)
r/3S'(r), as r > oo
(3.7)


59
5.1 Introduction
The Self-similar process is a realistic model for characterizing the statisti
cal behaviors of the traffic corresponding to LANs, WANs, and VBR coded video.
Self-similar processes are characterized by an hyperbolic decay of their autocorrela
tion function that cannot be captured by traditional Poisson-based processes. The
burstiness of multimedia traffic represents the main reason behind the development of
statistical multiplexing schemes for transport of heterogenous traffic over broadband
networks such as ATM networks. In ATM networks, cells are switched according to
information contained in their headers between ATM switches. When cells arrive
at a switch, they may be stored in a buffer awaiting transmission to their new des
tinations. If, within a certain period of time, the number of arriving cells, which
may come from different and often bursty sources, is larger than the number of cells
that can be served, the buffer may overflow and cells may be lost. In video services,
cell loss reduces picture quality. The design of flow and congestion control in ATM
networks is, therefore, faced with the challenge of how to provide required QoS and
still maintain sufficient network utilization to operate an economical viable networks.
Flow and congestion control procedures can be classified into preventive con
trol and reactive control with the former preventing the occurrence of congestion
with resource management, and the latter controlling the level of congestion based
on feedback information. Both approaches have advantages and disadvantages. In
ATM networks, a combination of these two approaches is used to provide effective
congestion control. For instance, best effort service uses a reactive scheme while CBR
and VBR are mainly based on preventive schemes.
In the previous chapter, we presented several reactive flow control schemes
such as rate-based or credit-based feedback flow control schemes. In contrast, a
preventive flow control generally involves the following two procedures: call admission
control (CAC), and bandwidth enforcement. Since ATM is a connection-oriented, a


21
To the Trunk
Interface Module
of other CPS 100
Switch
Figure 2.5: Functional model of the CPS-100 switch
interface module which consists principally of two controllers, namely an ingress adap
tor and an egress adaptor. The ingress adaptor receives cells from the incoming link,
buffers them as necessary, and sends them through the switching fabric. Similarly,
the egress adaptor accepts cells from the switching fabric, provides any necessary
buffering, and transmits the cells over the outgoing link. The breakdowns of the two
adaptors are shown in Figure 2.6. These adaptors also perform all the cell-by-cell
processing functions such as Virtual Connection Identifier (VCI) translation, priority
assignment and routing. Since the VCI is generally only local to each switch port,
the VCI of each cell must be translated to the value assigned for the succeeding links.
This operation is performed by a VCI translation table.
The ingress adaptor appends routing overhead on an incoming cell to specify
the output link associated with the virtual connection to which the cell belongs. The
ingress adaptor also looks up other connection related information such as the loss
priority and the service priority of the cell to assist the egress adaptor in making its
buffering and service scheduling decisions. Since all the cells of a virtual connection


31
1400
1 time (0.06 sec) 600
Figure 3.2: Self-similarity of WWW traffic



PAGE 1

()),&,(17 )/2: $1' &21*(67,21 &21752/ )25 6(/)6,0,/$5 75$)),& ,1 $70 1(7:25.6 %< :(1<(1 )8 $ ',66(57$7,21 35(6(17(' 72 7+( *5$'8$7( 6&+22/ 2) 7+( 81,9(56,7< 2) )/25,'$ ,1 3$57,$/ )8/),//0(17 2) 7+( 5(48,5(0(176 )25 7+( '(*5(( 2) '2&725 2) 3+,/2623+< 81,9(56,7< 2) )/25,'$

PAGE 2

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
PAGE 3

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

PAGE 4

6HOIVLPLODULW\ DQG +XUVW (IIHFW 6HOIVLPLODULW\ DQG 6ORZO\ 'HFD\LQJ 9DULDQFHV 0RGHOLQJ RI 6HOI6LPLODULW\ )UDFWLRQDO %URZQLDQ 0RWLRQ )UDFWLRQDO *DXVVLDQ 1RLVH )UDFWLRQDO $5,0$SGTf 3URFHVVHV 6HOI6LPLODULW\ 7KURXJK $JJUHJDWLRQ 6HOIVLPLODU 7UDIILF 7UDFH *HQHUDWRUV 5DQGRP 0LGSRLQW 'LVSODFHPHQW $OJRULWKP $JJUHJDWLRQ RI 5HQHZDO 3URFHVVHV 0*RR ZLWK +HDY\WDLOHG 'LVWULEXWHG 6HUYLFH 7LPH $5,0$ $SSUR[LPDWLRQ RI 3RZHU 6SHFWUXP &21*(67,21 $1' )/2: &21752/ *HQHUDO )UDPHZRUN IRU &RQJHVWLRQ 0DQDJHPHQW DQG &RQWURO )ORZ &RQWURO 6FKHPHV ZLWK )HHGEDFN &UHGLW%DVHG 6FKHPH 5DWH%DVHG 6FKHPH ,PSURSULHW\ RI 3RLVVRQ%DVHG 7UDIILF 0RGHOV 1HWZRUN &RQILJXUDWLRQ (IIHFWLYH 7KURXJKSXW &RPSDULVRQV &ODVVLFDO &RQWURO 7KHRU\ DQG &RQJHVWLRQ &RQWURO 35(9(17,9( )/2: &21752/ 6&+(0( )25 6(/)6,0,/$5 75$)n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

PAGE 5

&21&/86,21 $70 1HWZRUNV DQG 6HOIVLPLODULW\ 3UHYHQWLYH )ORZ &RQWURO )HHGEDFN )ORZ &RQWURO )XWXUH :RUN $33(1',&(6 $ 6&5,376 )25 *(1(5$7,1* 6(/)6,0,/$5 75$)),& $O 5DQGRP 0LGSRLQW 'LVSODFHPHQW $OJRULWKP $ 3RZHU 6SHFWUXP 'HQVLW\ $SSUR[LPDWLRQ ,OO % 6&5,376 )25 (67,0$7,1* +8567 3$5$0(7(5 %O 9DULDQFH 7LPH 3ORW $OJRULWKP % 5HVFDOHG $GMXVWPHQW $OJRULWKP 5()(5(1&(6 %,2*5$3+,&$/ 6.(7&+

PAGE 6

/,67 2) 7$%/(6 7HFKQRORJ\ &RPSDULVRQ 6HUYLFH &KDUDFWHULVWLFV RI $$/ /D\HUV 4R6 3DUDPHWHUV 3HUIRUPDQFH FRPSDULVRQ RI 3RLVVRQ SURFHVV DQG 663 WUDIILF PRGHOV 5REXVWQHVV RI WKH SURSRVHG &$& VFKHPH *XDUDQWHH RI 4R6 &/5 DQG QRUPDOL]HG XWLOL]DWLRQ UDWH XQGHU YDULRXV .Af &/5 DQG QRUPDOL]HG XWLOL]DWLRQ UDWH YV QXPEHU RI 9%5 9&V $UULYDO UDWH VWDWLVWLFV YL

PAGE 7

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

PAGE 8

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f 6HVVLRQ GLUHFWRU\ 6'5f 0XOWLFDVW VHVVLRQ ZLWK DXGLRYLGHR VWUHDPV 9LGHR DQG DXGLR VWUHDPV LQ D 0%RQH VHVVLRQ 1HWZRUN DSSOLFDWLRQV DQG 7&3,3 SURWRFRO VWDFN /,67B1(7 7RSRORJ\ 7UDFHV LQ /$1V DQG :$1V 9DULDQFH WLPH SORW 44SORW IRU 0ERQH WUDIILF 44SORW IRU 7&3 WUDIILF 44SORW IRU 8'3 WUDIILF 44SORW IRU RYHUDOO WUDIILF YLLL

PAGE 9

$EVWUDFW RI 'LVVHUWDWLRQ 3UHVHQWHG WR WKH *UDGXDWH 6FKRRO RI WKH 8QLYHUVLW\ RI )ORULGD LQ 3DUWLDO )XOILOOPHQW RI WKH 5HTXLUHPHQWV IRU WKH 'HJUHH RI 'RFWRU RI 3KLORVRSK\ ()),&,(17 )/2: $1' &21*(67,21 &21752/ )25 6(/)6,0,/$5 75$)),& ,1 $70 1(7:25.6 %\ :HQ
PAGE 10

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f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f FRQWUROOHU WR XSGDWH WKH RSWLPXP VHQGLQJ UDWH DW WKH WUDQVPLWWHU G\QDPLFDOO\ 7KH DGDSWLYH SROLF\ DWWHPSWV WR NHHS WKH EXIIHU RFFXSDQF\ IRU HDFK YLUWXDO FKDQQHO DW D VWHDG\ OHYHO DQG WKH VLPXODWLRQ UHVXOWV VKRZ WKDW WKH SURSRVHG VFKHPH ZRUNV HIIHFWLYHO\ DJDLQVW QHWZRUN FRQJHVWLRQ [

PAGE 11

&+$37(5 ,1752'8&7,21 $V\QFKURQRXV 7UDQVIHU 0RGH $70f LV RQH RI WKH PRVW SURPLVLQJ DQG ZLGHO\ UHFRJQL]HG VROXWLRQV IRU WKH KLJK VSHHG PXOWLPHGLD QHWZRUNV RI QH[W JHQHUDWLRQ 7KH DQWLFLSDWLRQ LV PDLQO\ GXH WR WKH VFDODEOH EXLOGLQJ EORFNV $70 SURYLGHV WR VHUYH DV D VWDEOH IRXQGDWLRQ IRU HYROYLQJ SXEOLF DQG SULYDWH QHWZRUNV $70 LV FDSDEOH RI KDQGOLQJ PDQ\ NLQGV RI LQIRUPDWLRQ LQFOXGLQJ YRLFH GDWD LPDJH DQG YLGHR LQ DQ LQWHJUDWHG PDQQHU DQG ZDV DFFHSWHG DV WKH WUDQVIHU PRGH IRU %URDGEDQG ,QWHJUDWHG 6HUYLFHV 'LJLWDO 1HWZRUNV %,6'1f E\ WKH ,QWHUQDWLRQDO 7HOHFRPPXQLFDWLRQV 8QLRQ ,78f LQ :LWK WKH DGYHQW RI %,6'1 $70 PXVW VXSSRUW DSSOLFDWLRQV ZLWK GLYHUVH WUDIILF FKDUDFWHULVWLFV DQG TXDOLW\ RI VHUYLFH 4R6f UHTXLUHPHQWV $Q DFFXn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

PAGE 12

5HFHQW VWXGLHV EDVHG RQ WKH WUDFH RI ORFDO DUHD QHWZRUN /$1f WUDIILF PDLQO\ DW %HOOFRUH >@ KDYH OHG WR WKH FRQFOXVLRQ WKDW /$1 WUDIILF FDQ QRW EH DGHTXDWHO\ UHSUHVHQWHG E\ WUDGLWLRQDO 0DUNRYEDVHG PRGHOV EXW LQVWHDG FDQ EH PRUH DSSURSULn DWHO\ PDWFKHG E\ VHOIVLPLODU RU IUDFWDOOLNH PRGHOV 0RUH UHFHQWO\ YDULDEOHELWUDWH 9%5f YLGHR WUDIILF > @ DQG ZLGH DUHD QHWZRUN :$1f WUDIILF >@ ZDV DOVR IRXQG WR H[KLELW VHOIVLPLODU FKDUDFWHULVWLFV :LWK WKH IRUHVHHDEOH SUHYDOHQFH RI ,QWHUQHW VHUYLFHV DQG PXOWLPHGLD DSSOLFDWLRQV $70 QHWZRUNV ZLOO EH FDUU\LQJ D VXEVWDQWLDO SRUWLRQ RI VHOIVLPLODU WUDIILF $V ZH NQRZ ZKHQ D QXPEHU RI EXUVW\ WUDIILF VRXUFHV DUH DFWLYH DQG QR LQWHUDFn WLYH RU SUHYHQWLYH DFWLRQ LV WDNHQ WR DJDLQVW LW FHOO ORVVHV WHQG WR RFFXU GXH WR EXIIHU RYHUIORZV 'HVSLWH WKDW UHVHDUFKHUV KDYH SURSRVHG PDQ\ IORZ DQG FRQJHVWLRQ FRQWURO VFKHPHV WR VROYH WKH SUREOHPV ZKHWKHU WKHVH DSSURDFKHV WDNH DGYDQWDJH RI WHFKn QLTXHV LQ IHHGEDFN FRQWURO > @ FDOO DGPLVVLRQ FRQWURO &$&f >@ WUDIILF VKDSLQJ DQG VPRRWKLQJ >@ RU UHVRXUFH UHVHUYDWLRQ DQG EXIIHU PDQDJHPHQW > @ QRQH RI WKHP WDNHV WKH VHOIVLPLODULW\ RI WKH WUDIILF LQWR FRQVLGHUDWLRQ $V D FRQVHn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

PAGE 13

'LVVHUWDWLRQ 2XWOLQH 7KLV GLVVHUWDWLRQ LV RUJDQL]HG DV IROORZV ,Q &KDSWHU ZH JLYH DQ LQWURGXFWLRQ RI $70 WHFKQRORJ\ DORQJ ZLWK YDULRXV WUDIILF FKDUDFWHULVWLFV LQ $70 QHWZRUNV :H GHVFULEH WKH $70 FHOO VWUXFWXUH LQ JUHDW GHWDLO 7KH ERWWRP WKUHH OD\HUV RI %,6'1 UHIHUHQFH PRGHO QDPHO\ 3K\VLFDO /D\HU $70 7UDQVSRUW /D\HU DQG $70 $GDSWDWLRQ /D\HU DUH SUHVHQWHG 7UDIILF FKDUDFWHULVWLFV RI &RQVWDQW %LWUDWH &%5f 9DULDEOH %LW UDWH 9%5f DQG $YDLODEOH %LWUDWH $%5f RU %HVWHIIRUW VHUYLFHV DQG 4R6 LQ $70 QHWZRUNV DUH GLVFXVVHG DW WKH HQG RI WKLV FKDSWHU ,Q &KDSWHU ZH ILUVW UHYLHZ WUDGLWLRQDO WUDIILF PRGHOV LQFOXGLQJ 3RLVVRQ 0DUNRY DQG 0DUNRY0RGXODWHG 3RLVVRQ PRGHO :H WKHQ IRFXV RQ VHOIVLPLODU SURn FHVVHV DQG WKHLU SURSHUWLHV DQG LPSOLFDWLRQ IRU QRYHO VHOIVLPLODU WUDIILF PRGHOV 7ZR RI WKH PRVW LPSRUWDQW SURSHUWLHV RI VHOIVLPLODU SURFHVVHV QDPHO\ +XUVW HIIHFW DQG ORQJ UDQJH GHSHQGHQFH ZLOO EH GLVFXVVHG DQG DQDO\]HG FRPSOHWHO\ :H LQWURGXFH YDULDQFHn WLPH SORWV 973f DQG UHVFDOHG DGMXVWHG 56f VWDWLVWLFV WR HVWLPDWH WKH +XUVW SDn UDPHWHU RI D VHOIVLPLODU VDPSOH SDWK :H SUHVHQW VHYHUDO V\QWKHVL]HG DOJRULWKPV IRU JHQHUDWLQJ VHOIVLPLODU VDPSOHV IRU H[DPSOH 5DQGRP 0LGSRLQW 'LVSODFHPHQW 50'f DOJRULWKP DQG ))7 DSSUR[LPDWLRQ ,Q WKLV UHVHDUFK PRVW DOJRULWKPV IRU HVn WLPDWLQJ WKH +XUVW SDUDPHWHU DQG JHQHUDWLQJ VHOIVLPLODU VDPSOH SDWK DUH FRQYHUWHG LQ 0DWODE VFULSWV ,Q &KDSWHU ZH LQWURGXFH D WUDIILF FRQWURO DQG PDQDJHPHQW IUDPHZRUN LQ $70 QHWZRUNV ,PSRUWDQW SDUWV RI WKH IUDPHZRUN DUH &RQQHFWLRQ $GPLVVLRQ &RQWURO &$&f 5HVRXUFH 0DQDJHPHQW 50f 7UDIILF 6KDSLQJ )HHGEDFN &RQWURO DQG VR RQ :H GHVFULEH VHYHUDO H[DPSOHV LQFOXGLQJ UDWHEDVWHG DQG FUHGLWEDVHG IORZ FRQWURO VFKHPHV 7KH SHUIRUPDQFH RI VHYHUDO FRQJHVWLRQ FRQWURO VFKHPHV DUH UHHYDOXDWHG E\ XVLQJ VHOIVLPLODU WUDIILF PRGHOV

PAGE 14

,Q &KDSWHU ZH SURSRVH D SUHYHQWLYH IORZ FRQWURO VFKHPH XVLQJ D &RQQHFWLRQ $GPLVVLRQ &RQWURO &$&f VXLWHG IRU JHQHULF VHOIVLPLODU WUDIILF SDUWLFXODUO\ PXOWLPHn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n WLPH 9%5 WUDIILF LQ $70 QHWZRUNV 7KH JRDO RI WKH VFKHPH LV WR PLQLPL]H WKH LPSDFW RI WUDIILF RYHUORDG LQ RUGHU WR OLPLW WKH FHOO ORVV UDWH WR DQ DFFHSWDEOH UDQJH DQG DOVR LQFUHDVH WKH QHWZRUN XWLOL]DWLRQ 7KH SURSRVHG IORZ FRQWURO VFKHPH LV EDVHG RQ SUHGLFWLQJ WKH HYROXWLRQ RI EXIIHU RFFXSDQF\ RYHU WLPH XVLQJ D 3URSRUWLRQDOSOXV ,QWHJUDOSOXV'HULYDWLYH 3,'f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n FRO 5693f 5HDOWLPH 7UDQVSRUW 3URWRFRO5HDOWLPH 7UDQVSRUW &RQWURO 3URWRFRO 57357&3f DQG ,3 0XOWLFDVWLQJ 3URWRFRO0%RQH 7R VWXG\ QHWZRUN G\QDPLFV DQG UHDO WUDIILF FKDUDFWHULVWLFV ZH FRQGXFWHG WUDIILF PHDVXUHPHQWV DQG DQDO\VLV EDVHG RQ HPSLULFDO WUDFHV FROOHFWHG LQ WKH /DERUDWRU\ IRU WKH ,QIRUPDWLRQ 6\VWHPV DQG 7HOHFRPPXQLFDWLRQV /,67f LQ $XJXVW :H LOOXVWUDWH WUDFHV VWDQGLQJ IRU PRUH

PAGE 15

WKDQ SDFNHWV LQ GLIIHUHQW WLPH VFDOHV 7KH WUDFHV DUH DOVR WUHDWHG VHSDUDWHO\ DFFRUGLQJ WR WKH SURWRFROV RI LQWHUHVW WR XV ZKLFK DUH ,3 0XOWLFDVWLQJ 7UDQVSRUW &RQWURO 3URWRFRO 7&3f DQG 8VHU 'DWDJUDP 3URWRFRO 8'3f $QG ILQDOO\ ZH FRQn FOXGH WKH GLVVHUWDWLRQ ZLWK VXPPDU\ DQG IXWXUH GLUHFWLRQV LQ &KDSWHU

PAGE 16

&+$37(5 %$&.*5281' $V\QFKURQRXV 7UDQVIHU 0RGH 7KH $70 VWDQGDUG LV SULPDULO\ GHILQHG E\ ,QWHUQDWLRQDO 7HOHJUDSK DQG 7HOHn SKRQH &RQVXOWDWLYH &RPPLWWHH &&,77f ODWHO\ FDOOHG ,QWHUQDWLRQDO 7HOHFRPPXQLFDn WLRQV 8QLRQ7HOHFRPPXQLFDWLRQ VWDQGDUGL]DWLRQ VHFWRU ,787f 6RPH LQWHULP VWDQn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n LQJ RI D ZHOO GHILQHG VHW RI IL[HG EDQGZLGWK FKDQQHOV (YHQWXDOO\ D VLGHHIIHFW RI WKLV SURYLVLRQ RI DUELWUDU\FDSDFLW\ FKDQQHOV $70 ZLOO DOVR EH DEOH WR VXSSRUW FKDQQHOV RI YDULDEOH ELW UDWH DQG KHQFH ZLOO EH FDSDEOH RI DFKLHYLQJ D VWDWLVWLFDO PXOWLSOH[LQJ JDLQ 60*f 7KLV VHUYLFH LV FXUUHQWO\ DYDLODEOH LQ VSHHGV UDQJLQJ IURP 0ESV 7 RU '6f WR 0ESV 2SWLFDO &DUULHU /HYHO RU 2&f ZLWK UDWHV H[SHFWHG WR EH DYDLODEOH GRZQ WR .ESV DQG XS WR 0ESV 2&f VRRQ 8OWLPDWHO\ $70 LV H[SHFWHG WR VFDOH XS WR WKH JLJDELW UDQJH $V D SURPLVLQJ FDQGLGDWH IRU QHWZRUNLQJ LQ /$1V DV ZHOO DV D UHSODFHPHQW IRU 7LPH 'LYLVLRQ 0XOWLSOH[LQJ 7'0f WUDQVPLVVLRQ V\VWHPV LQ :$1V UHVHDUFKHUV URXWLQHO\ VDZ $70 DV D VFDODEOH PHWKRG IRU WKH SURYLVLRQ RI KLJKVSHHG QHWZRUN

PAGE 17

$WWULEXWHV 'DWDFRP 7HOHFRP $70 7UDIILF 7\SH GDWD YRLFH GDWD YRLFH YLGHR 6ZLWFKLQJ SDFNHW FLUFXLW FHOO 8QLW YDULDEOH SDFNHW IL[HG IUDPH IL[HG FHOO 4XDOLW\ EHVW HIIRUW JXDUDQWHHG 4R6 $FFHVV WR 0HGLD VKDUHG GHGLFDWHG GHGLFDWHG &RQQHFWLRQ FRQQHFWLRQOHVV FRQQHFWLRQRULHQWHG FRQQHFWLRQRULHQWHG 7DEOH 7HFKQRORJ\ &RPSDULVRQ FRQQHFWLRQV WR URXWHUV DQG HQG V\VWHPV 7KLV WUHQG KDV EHHQ VR PDUNHG WKDW LW RYHUn WDNHV WKH GHYHORSPHQW RI $70 DV D EDFNERQH WUDQVPLVVLRQ PHWKRG DQG GLVSODFHV RWKHU GHYHORSPHQW SURVSHFWV IRU +LJK6SHHG /$1V DQG :$1V VXFK DV )UDPH 5HOD\ )5f )LEHU 'LVWULEXWHG 'LJLWDO ,QWHUIDFH )'',f DQG 6ZLWFKHG 0XOWLPHJDELW 'DWD 6HUYLFHV 60'6f $GGLWLRQDOO\ $70 LV VLJQLILFDQWO\ GLIIHUHQW IURP ERWK WKH SRSXODU GDWD FRPPXQLFDWLRQ DQG WHOHFRPPXQLFDWLRQ WHFKQRORJLHV LQ XVH WRGD\ $70 WHFKQROn RJ\ VWDQGV RXW E\ XQLI\LQJ WKHVH WZR ZRUOGV E\ ERUURZLQJ GHVLUDEOH DWWULEXWHV IURP ERWK GDWD FRPPXQLFDWLRQ DQG WHOHFRPPXQLFDWLRQ WHFKQRORJLHV >@ 7KH FRPSDULVRQ EHWZHHQ GDWD FRPPXQLFDWLRQ WHOHFRPPXQLFDWLRQ DQG $70 WHFKQRORJ\ LV VKRZQ LQ 7DEOH ([SOLFLW LQIRUPDWLRQ UHJDUGLQJ $70 ZLOO EH GLVFXVVHG LQ WKH QH[W IHZ VHFWLRQV $70 /D\HUHG 6WUXFWXUH 7KH $70 SURWRFRO XVHV D OD\HUHG VWUXFWXUH %,6'1 SURWRFRO UHIHUHQFH PRGHO VHH )LJXUH f VLPLODU WR WKH 2SHQ 6\VWHP ,QWHUFRQQHFW 26,f PRGHO +RZHYHU WKH WKUHHGLPHQVLRQDO OD\HU VWUXFWXUH RI $70 SURWRFRO LV GLVWLQFW IURP DOO RWKHU SURWRFRO PRGHOV LQ WKH XVH RI WKUHH SODQHV DFURVV DOO WKUHH $70 OD\HUV SOXV D KLJKHU OD\HU 7KH XVHU SODQH LV XVHG IRU HQGWRHQG GDWD WUDQVIHU WKH FRQWURO SODQH VXSSRUWV VLJQDOOLQJ XVHG WR HVWDEOLVKHG D FRQQHFWLRQ DQG WKH PDQDJHPHQW SODQH VXSSRUWV PDQDJHPHQW

PAGE 18

)LJXUH %,6'1 3URWRFRO UHIHUHQFH PRGHO DQG FRQWURO IXQFWLRQV IRU HDFK QHWZRUN DQG LWV HQGSRLQWV 7KH SULPDU\ OD\HUV RI WKH % ,6'1 SURWRFRO UHIHUHQFH PRGHO DUH WKH 3K\VLFDO /D\HU WKDW SURYLGHV IRU WUDQVPLVVLRQ RI $70 FHOOV RYHU D SK\VLFDO PHGLXP WKDW FRQQHFWV WZR $70 GHYLFHV WKH $70 /D\HU ZKHUH FHOO VHJPHQWDWLRQ DQG UHDVVHPEO\ RFFXUV DQG $70 $GDSWDWLRQ /D\HU $$/f WKDW SURYLGHV VXSSRUW IRU KLJKHU OD\HU VHUYLFHV VXFK DV FLUFXLW HPXODWLRQ IUDPH UHOD\ DQG 60'6 3K\VLFDO /D\HU 7KH 3K\VLFDO /D\HU RI WKH $70 VWDFN LV GLYLGHG LQWR WZR VXEOD\HUV WKH 7UDQVn PLVVLRQ &RQYHUJHQFH 7&f VXEOD\HU DQG WKH 3K\VLFDO 0HGLXP 30f VXEOD\HU 7KH 7& VXEOD\HU LV UHVSRQVLEOH IRU VXFK WDVNV DV OLQH FRGLQJ DQG ELW WLPLQJ WKH JHQHUDWLRQ DQG YHULILFDWLRQ RI WKH KHDGHU HUURU FRQWURO E\WH FHOO PDSSLQJ DQG FHOO GHOLQHDWLRQ DQG WKH 30 VXEOD\HU SURYLGHV ELWWUDQVPLVVLRQ FDSDELOLWLHV RQ DQ\ SDUWLFXODU PHGLXP LQFOXGLQJ RSWLFDO ILEHU FRD[LDO FDEOH 8QVKLHOGHG 7ZLVW 3DLU 873f DQG UDGLR OLQNV 8VHUV KDYH TXLHW D UDQJH RI 3K\VLFDO /D\HUV WR FKRRVH IURP 7KH $70 )RUXP >@ KDV LVVXHG UHFRPPHQGDWLRQV IRU 0SEV DQG 0ESV WUDQVPLVVLRQ RYHU PXOWL PRGH RSWLFDO ILEHU D 0ESV 6\QFKURQRXV 2SWLFDO 1HWZRUN 621(7f LQWHUIDFH D 0ESV 7 LQWHUIDFH DQG LV LQ WKH SURFHVV RI GUDZLQJ XS UHFRPPHQGDWLRQV IRU

PAGE 19

WUDQVPLVVLRQ RYHU XQVKLHOGHG WZLVWHG SDLU 873f 7KH &&,77,787 KDV FRQFHUQHG LWVHOI PDLQO\ ZLWK VWDQGDUGV VXLWDEOH IRU XVH ZLWK 6'+ 6\QFKURQRWLV 'LJLWDO +LHUDUn FK\f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f FODVV D YHFWRU RI WUDIILF SDUDPHWHUV 7KH $70 OD\HU WDNHV VWUHDPV RI E\WH FHOO SD\ORDGV DV LWV LQSXW IURP KLJKHU OD\HUV SHUIRUPV FHOO KHDGHU JHQHUDWLRQ DQG SDVVHV FHOOV WR WKH 7& VXEOD\HU RI WKH SK\VLFDO OD\HU VXFK WKDW RUGHU LV SUHVHUYHG ZLWKLQ YLUWXDO FKDQQHOV 9&Vf 7ZR OHYHOV RI 9&V FDQ EH VXSSRUWHG DW WKH XVHUQHWZRUN LQWHUIDFH 81,f WKDW FRQQHFWV FXVWRPHU SUHPLVHV HTXLSPHQW &3(f DQG $70 VZLWFKHV $ SRLQWWRSRLQW RU SRLQWWRPXOWLSRLQW 9LUWXDO &KDQQHO &RQQHFWLRQ 9&&f FRQVLVWLQJ RI D VLQJOH FRQQHFWLRQ HVWDEOLVKHG EHWZHHQ WZR $70 9&& HQG SRLQWV $ SRLQWWRSRLQW RU SRLQWWRPXOWLSRLQW 9LUWXDO 3DWK &RQQHFWLRQ 93&f FRQn VLVWLQJ RI D EXQGOH RI FRQQHFWLRQV HVWDEOLVKHG EHWZHHQ WZR $70 93& HQG SRLQWV 7KH $70 FHOO VWUXFWXUH LQ )LJXUH VKRZV WKH ILUVW ILYH E\WHV DUH DOORFDWHG WR WKH KHDGHU RI D IL[HG E\WH FHOO 7KH HQWLUH KHDGHU LV SURWHFWHG E\ D OE\WH +HDGHU (UURU &KHFN )(&f ILHOG :H QRWLFHG WKDW WKHUH LV QR UHWUDQVPLVVLRQ RI ORVW RU FRUUXSWHG GDWD SUHIRUPHG E\ WKLV OD\HU $Q DFFHVV IORZ FRQWURO PHFKDQLVP PD\

PAGE 20

_ E\WH ?r E\WH )LJXUH $70 &HOO )RUPDW DW 81, DQG 11, EH LPSOHPHQWHG DW WKH 81, DOWKRXJK FXUUHQWO\ QR FRQVHQVXV H[LVWV DV WR KRZ WKLV ZLOO ZRUN 7KLV DFFHVV IORZ FRQWURO ZLOO XVH WKH IRXU *HQHULF )ORZ &RQWURO *)&f ELWV DQG LV RQO\ PHDQLQJIXO ZKHQ WKHUH LV VKDUHG DFFHVV WR RQH $70 81, SRUW $W WKH UHFHLYLQJ HQG *)& UHFHLYHV DV LQSXW D VWUHDP RI FHOOV IURP WKH 3K\VLFDO OD\HU IURP ZKLFK LW SHUIRUPV KHDGHU H[WUDFWLRQ DQG GHOLYHUV FHOO LQ RUGHU WR WKH DSSURSULDWH $$/ VHUYLFH DFFHVV SRLQW 6$3f XVLQJ WKH 9LUWXDO &KDQQHO ,GHQWLILHU9LUWXDO 3DWK ,GHQWLILHU 9&,93,f YDOXHV DV LGHQWLILHUV $W VZLWFKLQJ HOHPHQWV WKH $70 OD\HU XVHV WKH 9&,93, WR URXWH WKH FHOOV :KLOH WKH 93, DQG 9&, YDOXHV PD\ FKDQJH DW HDFK VZLWFKLQJ HOHPHQW WKH $70 OD\HU GRHV WKLV WUDQVODWLRQ ,QWHUSUHWDWLRQ RI WKH YDOXHV LQ WKH 3D\ORDG 7\SHV 37f DQG &HOO/RVV 3ULRULW\ &/3f ILHOGV LV GRQH DW WKH $70 OD\HU LQ VZLWFKLQJ HOHPHQWV 7KH PDLQ IHDWXUH RI WKH 37 ,GHQWLILHU LV WR GLVFULPLQDWH XVHU FHOOV WKDW FDUU\ XVHU LQIRUPDWLRQ IURP QRQn XVHU FHOOV >@ 7KH 37 ,GHQWLILHU RI D XVHU FHOO PD\ QRW RQO\ LQGLFDWH ZKHWKHU WKH FHOO KDV H[SHULHQFHG FRQJHVWLRQ ZLWK ([SOLFLW )RUZDUG &RQJHVWLRQ ,QGLFDWLRQ ()&,f EXW DOVR VXJJHVW ZKHWKHU LW FRQWDLQV DQ LQGLFDWLRQ WR WKH $$/ SURWRFRO )RU H[DPSOH WKH Ef DW 11, , *)& 93, 93, 9&, 9&, 9&, 37 &/3 +(& 3$
PAGE 21

YDOXH f2LOf LQ WKH 37 ILHOG VXJJHVWV D XVHU FHOO IURP $$/ VHUYLFH FODVV DQG KDYLQJ H[SHULHQFHG FRQJHVWLRQ DQG WKH YDOXH ff VLPSO\ LQGLFDWHV D 5HVRXUFH 0DQDJHPHQW 50f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f DUH VLPLODU WKH GLIIHUHQFH EHLQJ WKDW WKH 81, ZLOO FRQQHFW DFFHVV SRUWV RI $70 VZLWFKHV DQG &XVWRPHU 3UHPLVH (TXLSPHQW &3(f ZKLFK FRXOG LQFOXGH EURDGEDQG WHUPLQDOV WHUPLQDO DGDSWRUV DQG FHOOEDVHG /$1:$1 HTXLSPHQW ZKLOH WKH 11, FDQ RQO\ FRQQHFW WUXQN SRUWV RI $70 VZLWFKHV 7KH 11, LV LQWHQGHG IRU $70 VXEQHWZRUNV RU QHWZRUNV DQG KHQFH GRHV QRW QHHG WKH *)& ILHOG LQ WKH $70 FHOO KHDGHU 0RUHRYHU WKH *)& ILHOG KDV EHHQ VXEVXPHG LQWR WKH 93, ILHOG DOORZLQJ WLPHV DV PDQ\ YLUWXDO SDWKV 7KHUH H[LVW D QXPEHU RI GLIIHUHQW 81,11, VSHFLILFDWLRQV HDFK RI ZKLFK LV EDVLFDOO\ D fVWDFNf FRPSULVLQJ RQH SDUWLFXODU SK\VLFDO OD\HU DQG WKH $70 OD\HU ,Q DGGLWLRQ WKH 81,11, HQFRPSDVVHV VLJQDOLQJ DQG FRQQHFWLRQ PDQDJHPHQW DVSHFWV IURP WKH FRQWURO SODQH $70 $GDSWDWLRQ /D\HU )URP SUHYLRXV VHFWLRQV WKH $70 OD\HU QRW RQO\ XVHV WKH VHUYLFH RI WKH SK\Vn LFDO OD\HU WR WUDQVSRUW FHOOV EXW DOVR GHOLYHUV FHOO SD\ORDGV WR WKH XSSHU OD\HUV ,Q PRVW FDVHV LW LV QHFHVVDU\ WR SHUIRUP VRPH DGDSWDWLRQ IXQFWLRQV IRU LQVWDQFH WR

PAGE 22

WUDQVIRUP D VHTXHQFH RI FHOOV LQWR WKH E\WH VWUHDP XVHG WR WUDQVSRUW GLJLWL]HG VLJQDOV 7KLV LV SHUIRUPHG E\ WKH $70 $GDSWDWLRQ /D\HU $$/f DV LOOXVWUDWHG LQ )LJXUH 7KH $70 DGDSWDWLRQ OD\HU $$/f FRQYHUWV WKH UHDO GDWD VWUHDP LQWR D FHOO VWUHDP DQG YLFH YHUVD *LYHQ WKH ZLGH YDULHW\ RI SRVVLEOH FKDUDFWHULVWLFV RI WKH LQSXW GDWD VWUHDPV RQH ZRXOG H[SHFW WR ILQG D YDULHW\ RI GLIIHUHQW DGDSWDWLRQ OD\HUV ,78 5HFn RPPHQGDWLRQ GHILQHV WKH EDVLF SULQFLSOHV DQG FODVVLILFDWLRQ RI $$/ IXQFWLRQV ZKLFK DUH GHVFULEHG DV IROORZV $$/ 6HUYLFHV &ODVVHV f &ODVV $ FRQVWDQW ELWUDWH &%5f VHUYLFH ZLWK HQGWRHQG WLPLQJ FRQQHFWLRQ RULHQWHG 7KH H[DPSOHV DUH IL[HGUDWH YLGHR DQG WKH FLUFXLW HPXODWLRQ VHUYLFHV VXFK DV '6O RU '6 WUDQVSRUW f &ODVV % YDULDEOH ELWUDWH 9%5f VHUYLFH ZLWK HQGWRHQG WLPLQJ FRQQHFWLRQ RULHQWHG 7KH H[DPSOHV DUH SDFNHWL]HG YRLFH DQG YLGHR f &ODVV & YDULDEOH ELWUDWH 9%5f VHUYLFH ZLWK QR HQGWRHQG WLPLQJ FRQQHFWLRQ RULHQWHG f &ODVV YDULDEOH ELWUDWH 9%5f VHUYLFH ZLWK QR HQGWRHQG WLPLQJ FRQQHFn WLRQOHVV $$/ )XQFWLRQV f $$/ LV GHVLJQHG WR WUDQVSRUW FRQQHFWLRQRULHQWHG &%5 GDWD VWUHDP LQ VXFK D ZD\ WKDW FORFN LQIRUPDWLRQ FDQ EH UHFRYHUHG DW WKH UHFHLYLQJ HQG WLPH WUDQVn SDUHQF\f $$/ LV LQ HIIHFW D YLUWXDO ZLUH 0RUHRYHU LW FDQ FRUUHFW VLQJOHELW HUURUV LQ WKH SD\ORDG DQG QRWLILHV ORVW FHOOV RU PLVRUGHUHG FHOOV f $$/ LV VSHFLILHG WR WUDQVSRUW FRQQHFWLRQRULHQWHG 9%5 GDWD VWUHDPV LQ VXFK D ZD\ WKDW WLPLQJ LQIRUPDWLRQ LV UHFRYHUDEOH DW WKH UHFHLYLQJ HQG $$/ KDV

PAGE 23

EHFRPH D NH\ SURWRFRO LQ $70 LPSOHPHQWDWLRQ UHTXLULQJ VXSSRUW IRU 9%5 DXGLR DQG YLGHR IRU H[DPSOH 0RWLRQ 3LFWXUH ([SHUW *URXS 03(*f YLGHR ,Q WKHRU\ WKLV $$/ FRXOG EH XVHG WR V\QFKURQL]H VHSDUDWH VWUHDPV RI GDWD ZLWKRXW WKH QHHG IRU HODERUDWH WLPHVWDPSLQJ LQ WKH GDWD VWUHDPV WKHPVHOYHV 'XH WR WKH VLPLODU WLPLQJ UHTXLUHPHQW DV $$/ $$/ FDQ DOVR FRUUHFW VLQJOHn ELW HUURUV LQ WKH SD\ORDG DQG QRWLILHV ORVW FHOOV RU PLVRUGHUHG FHOOV f $$/ LV WKH UHVXOW RI VWDQGDUGV HIIRUWV IRU WZR $$/V $$/ DQG $$/ ZKLFK DUH ERWK IRU WUDQVSRUW RI 9%5 VWUHDPV ZLWKRXW H[SOLFLW WLPLQJ LQIRUPDn WLRQ $$/ LV FRQQHFWLRQRULHQWHG DQG $$/ LV FRQQHFWLRQOHVV DOWKRXJK LW LV XQFOHDU ZKDW PHDQLQJ WKLV GLVWLQFWLRQ DFWXDOO\ KDV LQ $70 DQG XVXDOO\ WKHVH WZR VHUYLFHV DUH OXPSHG WRJHWKHU $V LW WXUQV RXW WKH WKHQf &&,77 PHUJHG $$/V DQG LQWR RQH VLQJOH W\SH ZKRVH SURFHGXUHV FDQ EH DSSOLHG LQ D FRQQHFWLRQRULHQWHG RU FRQQHFWLRQOHVV PDQQHU %RWK $$/ DQG $$/ SURYLGH DVVXUHG DQG QRQDVVXUHG VHUYLFHV ,Q WKH DVVXUHG PRGH WKH $$/ JXDUDQWHHV GHOLYHU\ RI $$/ 6HUYLFH 'DWD 8QLWV 6'8f LQ RUGHU DQG DQ\ ORVW 6'8V DUH UHWUDQVPLWWHG ,Q WKH QRQDVVXUHG PRGH WKLV IXQFWLRQ LV SHUIRUPHG E\ KLJKHU OD\HUV 7KH PDLQ DGYDQWDJH $$/ KDV LV WKDW LW FDUULHV D OHQJWK LQGLFDWLRQ LQ LWV ILUVW FHOO DQG KHQFH PDNH HDVLHU VRPH WUDIILF PDQDJHPHQW IXQFWLRQ VXFK DV IDVW EXIIHU UHVHUYDWLRQ f $$/ DOVR NQRZQ DV f6LPSOH DQG (IILFLHQW $GDSWDWLRQ /D\HU 6($/ff ZDV GHYHORSHG LQ UHVSRQVH WR D SHUFHSWLRQ WKDW $$/ ZDV LQHIILFLHQW VLQFH RQO\ RFWHWV RI D FHOO SD\ORDG FDUU\ DFWXDO GDWD $$/ LV D VXEVWDQWLDOO\ OHDQ $$/ FRPSDUHG ZLWK $$/ DW WKH H[SHQVH RI HUURU UHFRYHU\ DQG EXLOWLQ UHWUDQVPLVVLRQ 7KH WUDGHRII SURYLGHV D VPDOOHU EDQGZLGWK RYHUKHDG VLPSOHU SURFHVVLQJ UHTXLUHPHQWV DQG UHGXFHG LPSOHPHQWDWLRQ FRPSOH[LW\ ,Q $$/ WKHUH LV YHU\ OLWWOH $$/ RYHUKHDG ZLWK MXVW RQH KHDGHU DQG RQH ELW FKHFNVXP

PAGE 24

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n WLRQ ZLWK JXDUDQWHHG SHUIRUPDQFH 7KH SHUIRUPDQFH SDUDPHWHUV RU 4R6 FODVVHV DUH VXPPDUL]HG LQ 7DEOH $ 4R6 FODVV ZLWK D FRQQHFWLRQ FDQ KDYH HLWKHU VSHFLILHG RU XQVSHFLILHG SHUIRUPDQFH SDUDPHWHUV $ 6SHFLILHG 4R6 FODVV VSHFLILHV D VHW RI SHUn IRUPDQFH SDUDPHWHUV DQG WKH WDUJHW YDOXH IRU WKH FRUUHVSRQGLQJ SDUDPHWHU ,Q WKH

PAGE 25

$WWULEXWH $FFXUDF\ 6SHHG &HOO (UURU 5DWLR ; &HOO /RVV 5DWH ; &HOO 0LVLQVHUWLRQ 5DWH ; &HOO 7UDQVIHU 'HOD\ ; &HOO 'HOD\ 9DULDWLRQ ; 7DEOH 4R6 3DUDPHWHUV XQVSHFLILHG 4R6 FODVV QR YDOXH LV VSHFLILHG IRU WKH SHUIRUPDQFH SDUDPHWHUV HYHQ WKRXJK WKH QHWZRUN PD\ DVVLJQ WKRVH LQWHUQDO REMHFWLYHV 7KH SHUIRUPDQFH SDUDPHn WHUV DV GHILQHG LQ >@ FDQ EH IXUWKHU GLYLGHG LQ WZR FDWHJRULHV QDPHO\ DFFXUDF\ DQG VSHHG DV VKRZQ LQ 7DEOH $FFXUDF\ 7KH 4R6 FODVVHV UHJDUGLQJ DFFXUDF\ LQFOXGH &HOO (UURU 5DWLR &(5f &HOO /RVV 5DWH &/5f DQG &HOO 0LVLQVHUWLRQ 5DWH &05f 7KH &(5 IRU DQ $70 FRQQHFWLRQ LV GHILQHG DV &UU 1XPEHU RI HUURQHRXV FHOOV 7RWDO QXPEHU RI VXFFHVVIXO WUDQVPLWWHG DQG HUURQHRXV FHOOV f ZKHUH VXFFHVVIXO DQG HUURQHRXV FHOOV IRXQG LQ VHYHUHO\ IDXOW\ FHOOV EORFN DUH H[FOXGHG $ FHOO EORFN LV D VHTXHQFH RI FHOOV WUDQVPLWWHG FRQVHFXWLYHO\ RQ D FRQQHFWLRQ $ VHYHUHO\ IDXOW\ FHOO EORFN RFFXUV ZKHQ PRUH WKDQ D FHUWDLQ QXPEHU RI HUURU FHOOV ORVW FHOOV RU PLVLQVHUWHG FHOOV DUH REVHUYHG LQ D UHFHLYHG FHOO EORFN 7KH &/5 LV GHILQHG IRU DQ $70 FRQQHFWLRQ 9&& RU 93&f DV &(5 1XPEHU RI ORVW &HOOV 1XPEHU RI WUDQVPLWWHG &HOOV ,W LV QRWLFHG WKDW ORVW DQG WUDQVPLWWHG FHOOV IRXQG LQ VHYHUHO\ IDXOW\ EORFNV VKRXOG EH H[FOXGHG IURP WKH FHOO SRSXODWLRQ LQ FRPSXWLQJ &/5 7KH &05 IRU DQ 9&& RU 93& FRQQHFWLRQ LV GHILQHG DV Af 1XPEHU RI 0LVLQVHUWHG &HOOV &05 f§ LPH LQWHUYDO

PAGE 26

$JDLQ PLVLQVHUWHG FHOOV DSSHDULQJ LQ VHYHUHO\ IDXOW\ EORFNV VKRXOG EH H[FOXGHG &HOO PLVLQVHUWLRQ RQ D SDUWLFXODU FRQQHFWLRQ LV PRVW OLNHO\ FDXVHG E\ DQ XQGHWHFWHG HUURU LQ WKH KHDGHU RI D FHOO EHLQJ WUDQVPLWWHG RQ D GLIIHUHQW FRQQHFWLRQ 7KLV SHUIRUn PDQFH SDUDPHWHU LV GHILQHG DV D UDWH VLQFH WKH PHFKDQLVP SURGXFLQJ PLVLQVHUWHG FHOOV LV LQGHSHQGHQW RI WKH QXPEHU RI WUDQVPLWWHG FHOOV UHFHLYHG RQ WKH FRUUHVSRQGLQJ FRQQHFWLRQ 6SHHG 7KH 4R6 FODVVHV UHJDUGLQJ VSHHG LQFOXGH &HOO 7UDQVIHU 'HOD\ &7'f DQG &HOO 'HOD\ 9DULDWLRQ &'9f 7KH &7' LV GHILQHG DV WKH HODSVHG WLPH EHWZHHQ D FHOO H[LW IURP WKH VRXUFH DQG WKH FRUUHVSRQGLQJ FHOO DW GHVWLQDWLRQ 7KH &'9 LV DOVR FDOOHG fMLWWHUf 7KHUH DUH WZR SHUIRUPDQFH SDUDPHWHUV DVVRFLDWHG ZLWK &'9 SRLQW &'9 DQG SRLQW &'9 7KH SRLQW &'9 GHVFULEHV YDULDELOLW\ LQ WKH SDWWHUQ RI FHOO DUULYDO HYHQWV REVHUYHG DW D VLQJOH PHDVXUHPHQW SRLQW ZLWK UHIHUHQFH WR D QHJRWLDWHG SHDN UDWH 7KH SRLQW &'9 GHVFULEHV YDULDELOLW\ LQ WKH SDWWHUQ RI FHOO DUULYDO HYHQWV REVHUYHG DW WKH RXWSXW RI D FRQQHFWLRQ SRUWLRQ ZLWK UHIHUHQFH WR WKH SDWWHUQ RI WKH FRUUHVSRQGLQJ HYHQWV REVHUYHG DW WKH LQSXW WR WKH FRQQHFWLRQ SRUWLRQ 7UDIILF &KDUDFWHULVWLFV 1HWZRUN WUDIILF FDQ EH FKDUDFWHUL]HG LQ PDQ\ ZD\V GHSHQGLQJ RQ WKH DGDSWHG FULWHULRQ 2QH RI WKH PRVW ZLGHO\ XVHG FULWHULRQ LV WKH WROHUDQFH RI LQIRUPDWLRQ GHOD\ RU ORVV )RU LQVWDQFH UHDOWLPH WUDIILF VXFK DV WKDW IRU &%5 VHUYLFH LV GLVWLQJXLVKHG E\ LWV VWULQJHQW &7' DQG &'9 UHTXLUHPHQWV ZKHUHDV EHVW HIIRUW WUDIILF VXFK DV GDWD WUDQVPLVVLRQ GHPDQGV VXSHULRU &/5 EXW \LHOGV LQIHULRU SHUIRUPDQFH LQ WHUPV RI &7' DQG &'9 &RQVHTXHQWO\ DQ LPSURSHU FDOO DGPLVVLRQ FRQWURO DOJRULWKP FRXOG QRW RQO\ XQGHUXWLOL]H QHWZRUN UHVRXUFHV EXW DOVR PDNH GLIILFXOW WKH GHOLYHU\ RI WKH 4R6 RI HOLJLEOH XVHUV $QRWKHU H[DPSOH IRU WUDIILF FKDUDFWHULVWLFV LV WKDW VRPH &%5 WUDIILF VXFK DV YRLFH FDQ WROHUDWH D FHUWDLQ SRUWLRQ RI FHOO GLVFDUGLQJ LQ D FRQJHVWHG

PAGE 27

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f WKHQ VXFFHVVLYH FHOOV IURP WKH VDPH VRXUFH ZRXOG H[SHULHQFHV WKH VDPH GHOD\ RU ZLOO EH EORFNHGf XQWLO WKH VXSHUSRVHG DUULYDO SDWWHUQ XQGHUJRHV D FKDQJH 9%5 7UDIILF 7KH WUDIILF JHQHUDWHG E\ D W\SLFDO VRXUFH LQ JHQHUDO HLWKHU DOWHUQDWHV EHWZHHQ DFWLYH DQG VLOHQW SHULRGV DQGRU KDV D YDU\LQJ ELW UDWH JHQHUDWHG FRQWLQXRXVO\ )XUn WKHUPRUH WKH SHDNWRDYHUDJH ELW UDWH RI D 9%5 VRXUFH LV RIWHQ PXFK JUHDWHU WKDQ RQH 3UHVHQWLQJ 9%5 WUDIILF WR WKH QHWZRUN DV &%5 WUDIILF E\ PHDQV RI EXIIHU VPRRWKn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

PAGE 28

PRGHO SURYLGHV DQ DFFXUDWH DSSUR[LPDWLRQ RI WKH ELW UDWH RI D VLQJOH YLGHR VRXUFH ZLWKRXW VFHQH FKDQJHV EXW LV QRW TXLWH XVHIXO IRU TXHXHLQJ DQDO\VLV )LJXUH 79 &RPHUFLDO $%5 7UDIILF $YDLODEOH ELWUDWH $%5f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f LQWHUFRQQHFWLRQ

PAGE 29

)LJXUH 79 1HZV 7KH IDVW JURZWK LQ WKH QXPEHU RI SHUVRQDO FRPSXWHUV DQG ZRUNVWDWLRQV DQG WKH ULVLQJ QHHG IRU LQWHUFRQQHFWLQJ WKHP WRJHWKHU ZLWK IRU H[DPSOH VHUYHUV DQG SULQWHUV LQ DQ RIILFH KDV OHG WR WKH IDVW H[SDQVLRQ RI /$1V $V WKH QHHG IRU LQWHURIn ILFH FRPPXQLFDWLRQ URVH /$1V ZHUH LQ WXUQ LQWHUFRQQHFWHG YLD EULGJHV DQG URXWHUV &RQVHTXHQWO\ XVHUV VRRQ VWDUWHG WR GHPDQG VFDODEOH WKURXJKSXW IRU ODUJH YROXPHV RI GDWD H[FKDQJH EHWZHHQ WKHP ,Q DGGLWLRQ HPHUJLQJ PXOWLPHGLD DSSOLFDWLRQV WKDW LQWHJUDWH YRLFH GDWD LPDJH DQG YLGHR VWDUWHG WR FKDQJH WKH IDFH RI QHWZRUNLQJ LQ ORFDO DQG ZLGH DUHDV GXH WR WKHLU UHDO WLPH DQG KLJK EDQGZLGWK UHTXLUHPHQWV &36 ([DPSOH RI $70 6ZLWFKLQJ 'HVLJQ ,Q VSLWH RI WKH VWURQJ FRPSHWLWLRQ IURP WKH HYROYLQJ *LJDELW (WKHUQHW WHFKQROn RJ\ PRUH DQG PRUH UHVHDUFK DQG GHYHORSPHQW KDYH EHHQ LQYROYHG ZLWK $70 VZLWFKn LQJ WHFKQRORJ\ RYHU WKH \HDUV 7KH PDLQ UHDVRQ LV WKDW $70 VZLWFKHV KDYH SURYHG WR EH WKH EHVW RSWLRQ LQ WKH EDFNERQH RI ZLGH DUHD QHWZRUNV :$1Vf ,Q WKLV VHFWLRQ ZH SUHVHQW D IDVW SDFNHW VZLWFK DUFKLWHFWXUH EDVHG RQ D VKDUHG PHGLXP DQG RXWSXW EXIIHULQJ VFKHPH IRU ZKLFK VLPXODWLRQ DQG SHUIRUPDQFH VWXGLHV KDYH EHHQ FRQGXFWHG

PAGE 30

DW WKH 8QLYHUVLW\ RI )ORULGD >@ 7KH &36 VZLWFK LV DQ $70EDVHG VZLWFKLQJ V\VWHP WKDW VXSSRUWV $70 DQG YDULRXV GDWD VHUYLFHV VXFK DV 6ZLWFK 0HJDE\WH 'DWD 6HUYLFH 60'6f DQG )UDPH 5HOD\ IRU SXEOLF QHWZRUNV &36 VZLWFK LV D WLPH GLYLVLRQ VKDUHG PHGLXP VZLWFK IDEULF LQWHUFRQQHFWLQJ WLPH GLYLVLRQ VKDUHG PHPRU\ VZLWFKLQJ HOHPHQWV 7KH VZLWFK VWUXFWXUH LV JHQHUDOO\ FRPSRVHG RI WKH IROORZLQJ IRXU PDMRU FRPSRQHQWV 7KH VZLWFKLQJ IDEULF ZKLFK SHUIRUPV WKH EDVLF VZLWFKLQJ IXQFWLRQ RI WUDQVIHUULQJ FHOOV RU SDFNHWV IURP LQSXW WR RXWSXW 7KH VZLWFKLQJ IDEULF LQ WKH &36 VZLWFK LV FRPSRVHG RI D KLJKVSHHG 9LUWXDO $70 9$70f EDFNSODQH EXV DQG LV VKDUHG ZLWK DOO LQWHUIDFH PRGXOHV 7KH DFFHVV LQWHUIDFH PRGXOH SURYLGHV DQ LQWHUIDFH IRU FXVWRPHUV WR DFFHVV WKH VHUYLFHV VXSSRUWHG E\ QHWZRUN 7KH &XVWRPHU 3UHPLVHV (TXLSPHQW &3(f DWWDFKHV WR DQ DFFHVV LQWHUIDFH PRGXOH IRU WUDQVPLWWLQJ DQG UHFHLYLQJ GDWD YLD D GHGLFDWHG OLQN 7KH WUXQN LQWHUIDFH PRGXOH SURYLGHV DQ LQWHUIDFH IRU LQWHUFRQQHFWLQJ VZLWFKLQJ V\VWHPV WR VXSSRUW LQWHUVZLWFK FRPPXQLFDWLRQV 7KH &38 PRGXOH FRPPXQLFDWHV ZLWK DOO WKH LQWHUIDFH PRGXOHV LQ WKH VZLWFK DQG SHUIRUPV WKH KLJKOHYHO FRQWURO IXQFWLRQV VXFK DV FRQQHFWLRQ HVWDEOLVKPHQW DQG UHOHDVH PHPRU\ DGPLQLVWUDWLRQ EDQGZLGWK DOORFDWLRQ DQG PDLQWHQDQFH ,QWHUIDFH 0RGXOHV )LJXUH LOOXVWUDWHV WKH IXQFWLRQDO PRGHO RI D VL[SRUW &36 VZLWFK 1RWH WKDW DQ DFFHVVWUXQN LQWHUIDFH PRGXOH FDQ VXSSRUW PXOWLSOH ELGLUHFWLRQDO OLQNV %RWK H[WHUQDO WUXQNV DQG &3( OLQNV DUH FRQQHFWHG WR WKH 9$70 EXV YLD WKH DFFHVVWUXQN 7KH &36 ZDV GHYHORSHG E\ /RUDO 'DWD 6\VWHPV 6DUDVRWD )/

PAGE 31

7R WKH 7UXQN ,QWHUIDFH 0RGXOH RI RWKHU &36 6ZLWFK )LJXUH )XQFWLRQDO PRGHO RI WKH &36 VZLWFK LQWHUIDFH PRGXOH ZKLFK FRQVLVWV SULQFLSDOO\ RI WZR FRQWUROOHUV QDPHO\ DQ LQJUHVV DGDSn WRU DQG DQ HJUHVV DGDSWRU 7KH LQJUHVV DGDSWRU UHFHLYHV FHOOV IURP WKH LQFRPLQJ OLQN EXIIHUV WKHP DV QHFHVVDU\ DQG VHQGV WKHP WKURXJK WKH VZLWFKLQJ IDEULF 6LPLODUO\ WKH HJUHVV DGDSWRU DFFHSWV FHOOV IURP WKH VZLWFKLQJ IDEULF SURYLGHV DQ\ QHFHVVDU\ EXIIHULQJ DQG WUDQVPLWV WKH FHOOV RYHU WKH RXWJRLQJ OLQN 7KH EUHDNGRZQV RI WKH WZR DGDSWRUV DUH VKRZQ LQ )LJXUH 7KHVH DGDSWRUV DOVR SHUIRUP DOO WKH FHOOE\FHOO SURFHVVLQJ IXQFWLRQV VXFK DV 9LUWXDO &RQQHFWLRQ ,GHQWLILHU 9&,f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

PAGE 32

)URP /LQN $% 7R /LQN $ 7R /LQN % )LJXUH $GDSWHU ,QWHUIDFH RI D 6KDUHG 0HGLXP2XWSXW %XIIHULQJ $70 6ZLWFK

PAGE 33

ZLOO IROORZ WKH VDPH URXWH LQ WUDYHUVLQJ D VZLWFK WKH VHTXHQFH RI FHOOV LV SUHVHUYHG ZLWKRXW WKH QHHG RI H[SOLFLW VHTXHQFH QXPEHUV 7KH PRVW LPSRUWDQW IXQFWLRQ RI WKH HJUHVV DGDSWRU LV WR WHPSRUDULO\ EXIIHU WKRVH FHOOV ZKLFK FDQQRW EH GHOLYHUHG WR D UHFHLYHU LPPHGLDWHO\ $ ILQLWH EXIIHU PHPn RU\ DOORFDWHG LQ WKH HJUHVV DGDSWRU LV IRU UHVROYLQJ RXWSXW FRQWHQWLRQ ,I FRQJHVWLRQ FDXVHV WKH EXIIHUV WR ILOO XS WKH FRQWUROOHU KDV QR FKRLFH EXW WR GLVFDUG FHOOV 7KH EXIIHU WKXV IRUPV D FULWLFDO UHVRXUFH WKDW QHHGV WR EH DGHTXDWHO\ FRQWUROOHG $ PRGLILHG 3DUWLDO %XIIHU 6KDULQJ VFKHPH LV LPSOHPHQWHG LQ WKH HJUHVV DGDSn WRU WR HQVXUH VDWLVIDFWRU\ ORVV UDWHV IRU GLIIHUHQW ORVV SULRULW\ FODVVHV DQG IDLUQHVV LQ VKDULQJ WKH UHVRXUFH DPRQJ FRPSHWLQJ XVHUV 0RUHRYHU D SULRULW\GHSHQGHQW VHUn YLFH VFKHGXOLQJ VFKHPH LV UHDOL]HG WR FRQWURO WKH RUGHU LQ ZKLFK FHOOV DUH VHQW IURP WKH EXIIHU WR WKH RXWSXW OLQN 7KH VHUYLFH VFKHGXOLQJ FRQWURO LV QHFHVVDU\ IRU PHHWn LQJ WKH GLIIHUHQW VZLWFKLQJ GHOD\ UHTXLUHPHQWV IRU GLIIHUHQW VHUYLFH SULRULW\ FODVVHV 6HUYLFH DQG ORVV SULRULWLHV DUH DVVLJQHG WR HDFK YLUWXDO FRQQHFWLRQ GXULQJ FRQQHFWLRQ HVWDEOLVKPHQW EDVHG RQ WKH GHOD\ DQG ORVV UHTXLUHPHQWV VSHFLILHG IRU WKH SDUWLFXODU FRQQHFWLRQ ,W LV QRWLFHG WKDW RQO\ RQH LQJUHVV DGDSWRU FDQ KDYH DFFHVV WR WKH KLJKVSHHG EXV DW DQ\ WLPH 6R D EXV DUELWUDWLRQ VFKHPH LV LPSOHPHQWHG IRU UHVROYLQJ FRQWHQWLRQ IRU DFFHVV DPRQJ GLIIHUHQW LQJUHVV DGDSWRUV 6LQFH WKH VKDUHG PHGLXP RSHUDWHV DW D VXEVWDQWLDOO\ KLJKHU VSHHG DSSUR[LPDWHO\ *ESVf WKDQ WKDW RI OLQN LQWHUIDFH PRGn XOHV WKH &36 FDQ EH YLHZHG DV D QRQEORFNLQJ VZLWFK KDYLQJ QR LQWHUQDO EXIIHULQJ DQG ZLWK QR FRQWHQWLRQ YLVLEOH IURP RXWVLGH WKH VZLWFKLQJ IDEULF 7KRXJK WKH KLJK VSHHG RI WKH VZLWFKLQJ IDEULF UHGXFHV LQSXW FRQWHQWLRQ LW DOVR UHVXOWV LQ PDMRU TXHXHn LQJ WDNLQJ SODFH DW WKH EXIIHU RI HJUHVV DGDSWRUV 7KH SURSRVHG IORZ FRQWURO VFKHPH FDQ EH XVHG DORQJ ZLWK H[LVWLQJ EXIIHU PDQDJHPHQW VFKHPHV WR DYRLG LPPHQVH EXIIHU RYHUIORZV

PAGE 34

$GYDQWDJHV DQG /LPLWDWLRQV RI &36 6ZLWFK $ SULQFLSDO DGYDQWDJH RI WKH &36 $70 VZLWFK VWUXFWXUH LV LWV IOH[LELOn LW\ WR VXSSRUW PXOWLSOH NLQGV RI SRUW LQWHUIDFHV ,W KDV EHHQ ZLGHO\ UHFRJQL]HG WKDW $70 WHFKQRORJ\ LV WDUJHWHG WR VXSSRUW D ULFK PL[WXUH RI EURDGEDQG DQG QDUURZEDQG VHUYLFHV DV ZHOO DV WKH KLJKFDSDFLW\ LQWHUFRQQHFWLRQ RI H[LVWLQJ GDWD QHWZRUNLQJ DSn SOLFDWLRQV 7KXV LW LV LPSRUWDQW IRU D VZLWFKLQJ QRGH WR SURYLGH GLIIHUHQW NLQGV RI LQWHUIDFHV IRU WKH GLYHUVH DSSOLFDWLRQV LQ D KLJKO\ IOH[LEOH DQG H[SDQGDEOH ZD\ ,Q WKH &36 VZLWFK WKLV LV DFKLHYHG WKURXJK D PRGXODU GHVLJQ WKDW SODFHV WKH SRUWV RQ VHSDUDWH LQWHUIDFH PRGXOHV 9DULRXV FRPELQDWLRQV RI SRUW LQWHUIDFHV OLQN DFFHVV VSHHGV DQG SK\VLFDO PHGLD FDQ EH VXSSRUWHG HDVLO\ E\ LPSOHPHQWLQJ GLIIHUHQW LQWHUn IDFH PRGXOHV RQ WKH VZLWFK $ &36 VZLWFK FDQ EH FRQILJXUHG WR VXSSRUW XS WR RI WKHVH LQWHUIDFH PRGXOHV $QRWKHU DGYDQWDJH RI WKH &36 VZLWFK LV WKDW ZLWK WKLV DUFKLWHFWXUH G\n QDPLF SULRULW\ IXQFWLRQV DQG PXOWLFDVW RSHUDWLRQV FDQ EH VXSSRUWHG IOH[LEO\ ZLWK QR DGGLWLRQDO KDUGZDUH FRPSOH[LW\ 7KLV IXQFWLRQ LV SUHIHUDEOH VLQFH VRPH DSSOLFDWLRQV VXFK DV $70 /$1V UHTXLUH WKH DELOLW\ WR VXSSRUW PXOWLSOH JXDUDQWHHG FODVVHV RI VHUn YLFHV DQG IXOOEDQGZLGWK PXOWLFDVWLQJ (DFK LQWHUIDFH PRGXOH FDQ LPSOHPHQW VXLWDEOH EXIIHU PDQDJHPHQW DQG VFKHGXOLQJ SROLFLHV EDVHG RQ WKH WUDIILF FKDUDFWHULVWLF DV ZHOO DV ORDGLQJ FRQGLWLRQV WR EHVW VHUYH WKH GHOD\ DQG ORVV UHTXLUHPHQWV RI GLIIHUHQW WUDIILF FODVVHV ,Q DGGLWLRQ VLQFH WKH PDMRU TXHXHLQJ WDNHV SODFH DW WKH RXWSXW EXIIHU HDFK LQWHUIDFH PRGXOH FDQ SHUIRUP ORDG PRQLWRULQJ RQ WKH EXIIHU XVDJHV RI GLIIHUHQW WUDIILF FODVVHV WR VXSSRUW FRQJHVWLRQ FRQWURO RSHUDWLRQ /DVWO\ WKH EXIIHU VSDFH ZLWKLQ DQ LQWHUIDFH PRGXOH FDQ EH DOORFDWHG G\QDPLFDOO\ WR HDFK SRUW VR DV WR DFKLHYH D ORZHU SUREDELOLW\ RI PHPRU\ RYHUIORZ 2Q WKH RWKHU KDQG DV ZLWK RWKHU VKDUHG PHGLXPRXWSXW EXIIHULQJ $70 VZLWFKHV WKH &36 VZLWFK DUFKLWHFWXUH KDV WKH W\SLFDO OLPLWDWLRQ RI UHTXLULQJ WKH

PAGE 35

KLJK VSHHG VZLWFKLQJ IDEULF WR DFFRPPRGDWH VLPXOWDQHRXV DUULYDOV IURP GLIIHUHQW LQn SXW SRUWV 7KH VZLWFKLQJ IDEULF DQG RXWSXW EXIIHU PXVW HIIHFWLYHO\ RSHUDWH DW D PXFK KLJKHU VSHHG WKDQ WKDW RI HDFK LQWHUIDFH SRUW 7KH LPSOHPHQWDWLRQ RI WKH KLJKVSHHG EXV DQG EXIIHU PHPRU\ FRXOG EH FRPSOH[ LI WKH UHTXLUHG PHPRU\ DFFHVV VSHHG LV YHU\ KLJK WKXV OLPLWLQJ WKH FDSDELOLW\ RI WKH VZLWFK WR VXSSRUW YHU\ KLJK VSHHG LQWHUIDFH SRUWV

PAGE 36

&+$37(5 75$)),& 02'(//,1* $1' 6(/)6,0,/$5 352&(66(6 $ UHFHQWO\ REVHUYHG SKHQRPHQRQ UHODWHG WR WUDIILF LQ %URDGEDQG ,QWHJUDWHG 6HUYLFH 'LJLWDO 1HWZRUNV %,6'1f LV WKH fVHOIVLPLODULW\f RU fEXUVWLQHVVf H[KLELWHG E\ NH\ VHUYLFHV VXFK DV FRPSUHVVHG YLGHR >@ ILOH WUDQVIHU >@ DQG ::: >@ 7ZR PDLQ VRXUFHV RI EXUVWLQHVV DUH GXH WR WKH VKDSHV RI WKH PDUJLQDO GLVWULEXWLRQ DV ZHOO DV WKH DXWRFRUUHODWLRQ IXQFWLRQ RI SDFNHW DUULYDO UDWH RU LQWHUDUULYDO WLPHV >@ :KHQ PRGHOLQJ QHWZRUN WUDIILF SDFNHW DQG FRQQHFWLRQ DUULYDOV DUH RIWHQ DVn VXPHG WR EH 3RLVVRQ SURFHVVHV EHFDXVH VXFK SURFHVVHV KDYH DWWUDFWLYH WKHRUHWLFDO SURSHUWLHV $ QXPEHU RI VWXGLHV KDYH VKRZQ KRZHYHU WKDW IRU ERWK ORFDODUHD DQG ZLGHDUHD QHWZRUN WUDIILF WKH GLVWULEXWLRQ RI SDFNHWV LQWHU DUULYDO WLPH FOHDUO\ GLIIHUV IURP H[SRQHQWLDO > @ 5HFHQW ZRUNV DUJXH FRQYLQFLQJO\ WKDW /$1 WUDIILF LV PXFK EHWWHU PRGHOHG XVLQJ VWDWLVWLFDOO\ VHOIVLPLODU SURFHVVHV ZKLFK KDYH VLJQLILFDQWO\ GLIIHUHQW WKHRUHWLFDO SURSHUWLHV WKDQ 3RLVVRQ SURFHVVHV > @ $Q DSSURn SULDWH WUDIILF PRGHOLQJ ZKLFK UHIOHFWV WKH EXUVWLQHVV RU VHOIVLPLODULW\ LV KHQFH FULWLFDO WR WKH VXFFHVV RI SHUIRUPDQFH HYDOXDWLRQ 7KH UHPDLQGHU RI WKLV FKDSWHU LV RUJDQL]HG DV IROORZV &RPPRQO\ XVHG 3RLVVRQ DQG 3RLVVRQEDVHG WUDIILF PRGHOV DUH SUHVHQWHG LQ 6HFWLRQ 7KH VHOIVLPLODU SURFHVVHV DQG WKHLU SURSHUWLHV DUH JLYHQ LQ 6HFWLRQ 6HFWLRQ GHVFULEHV WKH PRGHOLQJ RI 6HOIn VLPLODU SURFHVVHV DQG ILQDOO\ VHYHUDO SRSXODU VHOIVLPLODU SURFHVVHV JHQHUDWRUV DUH GHVFULEHG LQ 6HFWLRQ

PAGE 37

3RLVVRQ%DVHG 7UDIILF 0RGHOV 3RLVVRQ 3URFHVV 0RGHOV 3RLVVRQ SURFHVV PRGHOV DUH WKH ROGHVW WUDIILF PRGHOV GDWLQJ EDFN WR WKH DGn YHQW RI WHOHSKRQ\ SULPDULO\ FRQWULEXWHG E\ WKH ZHOONQRZQ SLRQHHULQJ HQJLQHHU $ (UODQJ LQ WKH V ,W LV ZHOO NQRZQ WKDW WKH WUDIILF RQ WHOHFRPPXQLFDWLRQ QHWZRUNV FDQ EH PRGHOOHG DQG FKDUDFWHUL]HG E\ D 3RLVVRQ SURFHVV 2QH RI WKH PRVW LPSRUWDQW WUDIILF FKDUDFWHULVWLFV RQ WHOHFRPPXQLFDWLRQ QHWZRUNV LV WKDW PRVW KXPDQ LQLWLDWHG SURFHVVHV DUH PHPRU\OHVV $ 3RLVVRQ SURFHVV FDQ EH FKDUDFWHUL]HG DV D UHQHZDO SURn FHVV ZKRVH LQWHUDUULYDO WLPHV ^f` DUH H[SRQHQWLDOO\ GLVWULEXWHG ZLWK UDWH RI $ WKDW LV 3U^f W` f§ H[Sf§$Lf f (TXLYDOHQWO\ LW LV D FRXQWLQJ SURFHVV VDWLVI\LQJ 3U^$Lf f§ Qf fS f H[Sf§$Wf f ZKHUH $>Wf LV WKH QXPEHU RI DUULYDOV LQ D WLPH LQWHUYDO Wf 3RLVVRQ SURFHVVHV HQMR\ VRPH HOHJDQW DQDO\WLFDO SURSHUWLHV VXFK DV 7KH QXPEHU RI DUULYDOV LQ GLVMRLQW LQWHUYDOV DUH LQGHSHQGHQW ZKLFK LV UHIHUUHG WR DV fLQGHSHQGHQW LQFUHPHQWf SURSHUW\ 0HUJLQJ LQGHSHQGHQW 3RLVVRQ SURFHVVHV UHVXOW LQ D QHZ 3RLVVRQ SURFHVV ZLWK D UDWH EHLQJ WKH VXP RI WKH FRPSRQHQW UDWHV 7KH LQGHSHQGHQW LQFUHPHQW SURSHUW\ UHQGHUV 3RLVVRQ D PHPRU\OHVV SURFHVV 7KLV LQ 3LUQ JUHDWO\ VLPSOLILHV TXHXHLQJ SUREOHPV LQYROYLQJ 3RLVVRQ DUULYDOV ,Q VSLWH RI WKH DQDO\WLF VLPSOLILFDWLRQ 3RLVVRQ SURFHVVHV GR KDYH D VLJQLILFDQW PRGHOLQJ

PAGE 38

GUDZEDFN WKH LQFDSDFLW\ RI FDSWXULQJ WKH DXWRFRUUHODWLRQ IXQFWLRQ RI ^,Q` ZKLFK WR D ODUJH H[WHQW H[SODLQV WKH SKHQRPHQRQ RI WUDIILF EXUVWLQHVV 0DUNRY 3URFHVV 0RGHOV 8QOLNH 3RLVVRQ SURFHVV PRGHOV 0DUNRY SURFHVV PRGHOV LQWURGXFH GHSHQGHQFH LQWR WKH UDQGRP VHTXHQFH ^,Q` VR WKDW WKH\ FDQ SRWHQWLDOO\ FDSWXUH WUDIILF EXUVWLQHVV &RQVLGHU D FRQWLQXRXVWLPH 0DUNRY SURFHVV ;Wf ZLWK D GLVFUHWH VWDWH VSDFH L e ^ P` &RQVHTXHQWO\ D VHW RI UDQGRP YDULDEOHV ; ^;L` IRUP D 0DUNRY FKDLQ ZKHUH WKH GHSHQGHQF\ H[WHQGV EDFNZDUGV RQO\ RQH XQLW LQ WLPH ,Q RWKHU ZRUGV WKH ZD\ LQ ZKLFK WKH HQWLUH SDVW KLVWRU\ DIIHFWV WKH IXWXUH RI WKH SURFHVV LV FRPSOHWHO\ VXPPDUL]HG LQ WKH FXUUHQW VWDWH RI WKH SURFHVV 7KLV SURSHUW\ LV VR FDOOHG f0DUNRYf RU fPHPRU\OHVVf SURSHUW\ )LJXUH LOOXVWUDWHV DQ H[DPSOH RI VWDWH 0DUNRY FKDLQ ZKHUH ; VWD\V LQ D VWDWH IRU DQ H[SRQHQWLDOO\ GLVWULEXWHG KROGLQJ WLPH UHTXLUHG WR VDWLVI\ WKH 0DUNRY SURSHUW\f ZLWK SDUDPHWHU $R DQG WKHQ MXPSV WR VWDWH ZLWK SUREDELOLW\ D ,Q VWHDG\ VWDWH ZH KDYH WKH VWDWH HTXDWLRQ 3R 3L 4 f§ D D f§ c 3R 3L 3R 3L f ZKHUH SR DQG S? DUH WKH VWHDG\VWDWH SUREDELOLW\ RI ;T DQG ;? UHVSHFWLYHO\ DQG 4 LV FDOOHG D WUDQVLWLRQ SUREDELOLW\ PDWUL[ 1RWH WKDW LQ D VLPSOH 0DUNRY WUDIILF PRGHO HDFK MXPS RI WKH 0DUNRY SURFHVV ; LV LQWHUSUHWHG DV VLJQDOLQJ DQ DUULYDO VR LQWHUDUULYDO WLPHV DUH H[SRQHQWLDOO\ GLVWULEXWHG ZLWK WKHLU UDWH SDUDPHWHUV GHSHQGHQW RQ WKH VWDWH IURP ZKLFK WKH MXPS RFFXUUHG 7KLV VLWXDWLRQ UHVXOWV LQ GHSHQGHQFH DPRQJ LQWHUDUULYDO WLPHV DV D FRQVHTXHQFH RI WKH 0DUNRY SURSHUW\

PAGE 39

D 3 )LJXUH VWDWH 0DUNRY FKDLQ 0DUNRY0RGXODWHG 3URFHVV 0RGHOV 0DUNRYPRGXODWHG SURFHVV PRGHOV FRQVWLWXWH DQ H[WUHPHO\ LPSRUWDQW FODVV RI WUDIILF PRGHOV 7KH LGHD LV WR LQWURGXFH DQ H[SOLFLW QRWLRQ RI VWDWH LQWR WKH GHVFULSWLRQ RI D WUDIILF VWUHDP DQ DX[LOLDU\ 0DUNRY SURFHVV LV HYROYLQJ LQ WLPH DQG LWV FXUUHQW VWDWH FRQWUROV PRGXODWHVf WKH SUREDELOLW\ ODZ RI WKH WUDIILF PHFKDQLVP 1RZ DVVXPH WKDW ZKLOH ; LV LQ VWDWH L WKH SUREDELOLW\ GLVWULEXWLRQ IXQFn WLRQ SGIf RI DUULYDOV LV FRPSOHWHO\ GHWHUPLQHG E\ L DQG WKLV KROGV IRU HYHU\" ^ P` 1RWH WKDW ZKHQ ; XQGHUJRHV D WUDQVLWLRQ WR VWDWH M f ^ P` WKHQ D SGI IRU DUULYDOV WDNHV HIIHFW IRU WKH GXUDWLRQ RI VWDWH M DQG VR RQ 7KXV WKH SGI IRU DUULYDOV LV PRGXODWHG E\ WKH VWDWH RI ; 7KH PRVW FRPPRQO\ XVHG 0DUNRYPRGXODWHG SURFHVV LV 0DUNRY0RGXODWHG 3RLVVRQ 3URFHVV 0033f ZKLFK FRPELQHV WKH VLPSOLFLW\ RI WKH PRGXODWLQJ 0DUNRYf SURFHVV ZLWK WKDW RI WKH PRGXODWHG 3RLVVRQf SURFHVV ,Q WKLV FDVH WKH PRGXODWLRQ PHFKDQLVP VLPSO\ VWLSXODWHV WKDW LQ VWDWH L RI ; DUULYDOV RFFXU DFFRUGLQJ WR D 3RLVn VRQ SURFHVV DW UDWH $$V WKH VWDWH FKDQJHV VR GRHV WKH UDWH $V D VLPSOH H[DPSOH FRQVLGHU D WZRVWDWH 0033 PRGHO ZKHUH RQH VWDWH LV DQ fRQf VWDWH ZLWK DQ DVVRFLn DWHG SRVLWLYH 3RLVVRQ UDWH DQG WKH RWKHU LV DQ fRIIf VWDWH ZLWK DVVRFLDWHG UDWH ]HUR VXFK PRGHOV DUH DOVR NQRZQ DV LQWHUUXSWHG 3RLVVRQf 7KHVH PRGHOV KDYH EHHQ ZLGHO\

PAGE 40

XVHG WR PRGHO YRLFH WUDIILF VRXUFHV ZKHUH WKH fRQf VWDWH FRUUHVSRQGV WR WDON VSXUW DQG WKH fRIIf VWDWH FRUUHVSRQGV WR VLOHQFH 6HOIVLPLODU 3URFHVVHV 2QH RI WKH PRVW UHPDUNDEOH DVSHFWV DERXW VHOIVLPLODULW\ LQ fV LV SUREDEO\ WKDW LW KDV EHHQ FRQVLVWHQWO\ REVHUYHG LQ UHFHQW VWXGLHV DQG PHDVXUHPHQWV LQ QHWZRUN WUDIILF > @ 6HOIVLPLODULW\ LV PDQLIHVWHG LQ VXFK D ZD\ WKDW WKHUH LV QR QDWXUDO OHQJWK IRU D fEXUVWf DFURVV DOO WLPH VFDOHV ,Q RWKHU ZRUGV D EXUVWV FRQVLVWV RI EXUVW\ VXESHULRGV VHSDUDWHG E\ OHVV EXUVW\ VXESHULRGV )RU LQVWDQFH IUDFWDO LPDJHV LQ FRPSXWHU JUDSKLFV DQG VQRZIODNHV DUH SUREDEO\ WZR RI WKH PRVW ZHOONQRZQ REMHFWV H[KLELWLQJ VHOIVLPLODU RU fIUDFWDOOLNHf SKHQRPHQD ZKHUH HDFK VPDOO SRUWLRQ RI WKHP FDQ EH YLHZHG DV D UHGXFHGVFDOH GXSOLFDWH RI WKH ZKROH 6RPH DVSHFWV RI fVHOI VLPLODULW\f FRLQHG E\ 0DQGHOEURW >@ DOVR DSSHDUV LQ K\GURORJ\ HFRQRPLFV DQG FRPPXQLFDWLRQV DQG LV UHODWHG WR VR FDOOHG fI QRLVHVf >@ 6HOIVLPLODU WUDIILF LQ QHWZRUNV EHKDYHV VR GLIIHUHQWO\ IURP YRLFH WUDIILF WKDW QRQH RI WKH WUDGLWLRQDO 3RLVVRQEDVHG WUDIILF PRGHOV GHVFULEHG LQ 6HVVLRQ LV DEOH WR FDSWXUH WKH IUDFWDO EHKDYLRUV ,W LV DOVR VKRZQ WKDW WKH JHQHUDOO\ DFFHSWHG DUJXPHQW WKDW DJJUHJDWH WUDIILF EHFRPHV VPRRWKHU DV WKH QXPEHU RI WUDIILF VRXUFHV LQFUHDVH KDV YHU\ OLWWOH WR GR ZLWK WKH UHDOLW\ LQ (WKHUQHW /$1V >@ 2Q WKH FRQWUDU\ WKH EXUVWLQHVV RI /$1 WUDIILF W\SLFDOO\ LQWHQVLILHV DV WKH DPRXQW RI DFWLYH WUDIILF LQFUHDVHV )LJXUH LOOXVWUDWHV WKH HPSLULFDO WUDFH FROOHFWHG IURP D ::: VHUYHU DW D W\SLFDO ,QWHUQHW VHUYLFH SURYLGHU ,63f LQ -XQH 7KH EUHDNGRZQV RI SDFNHW DUULYDO UDWH DUH SORWWHG DFFRUGLQJ WR WLPH VFDOHV UDQJLQJ IURP VHFRQG WR VHFRQGV 2QH FDQ REVHUYH WKH VLPLODU DSSHDUDQFH RI fEXUVWLQHVVf DFURVV D ZLGH UDQJH RI WLPH VFDOHV ZKHUH WKH SORW RQ WKH WRS FRUUHVSRQGV WR WKH FRPSOHWH WUDFH ZLWK D UHVROXWLRQ RI VHFRQGV WKH VHFRQG IURP WKH WRS DQ H[WUDFWLRQ RI WKH ILUVW RQH OWKf EXW ZLWK WHQ WLPHV ILQHU UHVROXWLRQ VHFRQGVf DQG VR RQ ,W LV HYLGHQW WKDW WKH EXUVWLQHVV SHUVLVWV RYHU GLIIHUHQW WLPHVFDOHV DQG WKDW WKH WUDIILF LV VWDWLVWLFDOO\ VHOIVLPLODU

PAGE 41

WLPH VHFf )LJXUH 6HOIVLPLODULW\ RI ::: WUDIILF

PAGE 42

'HILQLWLRQV $ ZLGHVHQVH VWDWLRQDU\ :66f UDQGRP SURFHVV ;Wf LV GHILQHG DV H[DFWO\ VHOIn VLPLODU >@ LI D VPDOO SRUWLRQ RI ;Wf FDQ H[DFWO\ UHSURGXFH D ODUJH SRUWLRQ RI ;Wf E\ VFDOLQJ SURSHUO\ :KDW LV RI PRUH LQWHUHVW WR QHWZRUN SHUIRUPDQFH DQDO\VLV KRZHYHU LV WKH VR FDOOHG VWDWLVWLFDO VHOIVLPLODU >@ $ :66 UDQGRP SURFHVV ;Wf LV VWDWLVWLFDOO\f VHOIVLPLODU LI HDFK VPDOOHU SRUWLRQ VKDUHV VWDWLVWLFDO SURSHUWLHV RI ODUJHU SRUWLRQ RI ;Wf DIWHU SURSHUO\ VFDOHG $ PRUH H[SOLFLW GHILQLWLRQ RI VHOIVLPLODU LQ >@ LV GHILQHG EHORZ 'HILQLWLRQ ;Wf LV VHOIVLPLODU ZLWK f+XUVW SDUDPHWHUf + @ LI ; Wf ?D?a+ ; DWf 9D f DQG ;W Uf ;Wf ;Uf ;f 9 U f ZKHUH GHQRWHV WKH HTXLYDOHQFH RI GLVWULEXWLRQ %\ WKLV GHILQLWLRQ D VHOIVLPLODU SURFHVV KDV VWDWLRQDU\ LQFUHPHQWV DV LQGLFDWHG LQ f ,Q DGGLWLRQ WKLV SURFHVV LV DOVR GLVWULEXWLRQ LQYDULDQW XQGHU DSSURSULDWH VFDOn LQJ LQ WLPH DQG VSDFH LQ D VHQVH WKDW WKH VHOIVLPLODULW\ FDQ EH REVHUYHG E\ FRPSDULQJ VPDOO EXUVWV LQ D VPDOO UHJLRQ ZLWK ODUJHU RQHV ZLWK VPDOOHU RQHV VXSHULPSRVHG RYHU D ODUJHU UHJLRQ RI WKH KRUL]RQWDO D[LV 7KLV GLVWULEXWLRQ LQYDULDQW FKDUDFWHULVWLF LV REVHUYHG LQ DQ HPSLULFDO WUDFH VKRZ LQ )LJXUH DV ZHOO 7KH YDULDWLRQV RI WKH GHILn QLWLRQ RI VHOIVLPLODULW\ ZKLFK OHDG WR PRUH LPSRUWDQW SURSHUWLHV ZLOO EH GHVFULEHG DV IROORZV /HW ;Wf EH D :66 UDQGRP SURFHVV ZLWK PHDQ L YDULDQFH D DQG DXWRFRUUHn ODWLRQ IXQFWLRQ UUf ,Q SDUWLFXODU ZH DVVWLPH WKDW UUf LV RI WKH IRUP W +6Wf DV U f§! RR f U6nUf DV U f§! RR f

PAGE 43

ZKHUH 2 + DQG 6Wf LV VORZO\ YDU\LQJ DW LQILQLW\ QDPHO\ OLPARR 6W[f6Wf IRU DOO [ 'HQRWH :Pf WKH QHZ SURFHVV REWDLQHG E\ DYHUDJLQJ WKH RULJLQDO VHULHV ; LQ QRQRYHUODSSLQJ VXE EORFNV RI VL]H P 7KDW LV Pf§ ;:Nf < ;NPL f L 1RWH WKDW IRU HDFK P GHILQHV D :66 UDQGRP SURFHVV 'HILQLWLRQ 'HQRWH UAP?7f WKH DXWRFRUUHODWLRQ IXQFWLRQ RI ;AP? 7KH SURFHVV ;Wf LV fH[DFWO\ VHFRQGRUGHU VHOIVLPLODUf >@ ZLWK +XUVW SDUDPHWHU + LI UAUf UUf IRU DOO P f ,Q RWKHU ZRUGV ; LV H[DFWO\ VHFRQGRUGHU VHOIVLPLODU LI WKH DJJUHJDWHG SURn FHVVHV DUH LQGLVWLQJXLVKDEOH IURP ; ZLWK UHVSHFW WR WKHLU ILUVW DQG VHFRQG RUGHU SURSn HUWLHV :LWK WKH UHOD[DWLRQ RI f ZH DUULYH DW WKH IROORZLQJ GHILQLWLRQ 'HILQLWLRQ ; LV fDV\PSWRWLFDOO\ VHOIVLPLODUf ZLWK +XUVW SDUDPHWHU + LI UAP?Uf f >U Of+ U f§ ?f+ f§ U+@ DV P f§! RR f 6SHFLILFDOO\ IRU ODUJH P WKH DJJUHJDWHG SURFHVV ;A KDV D IL[HG DXWRFRUUHODn WLRQ VWUXFWXUH GHWHUPLQHG RQO\ E\ + 7KH PRVW VWULNLQJ IHDWXUH RI VHOIVLPLODULW\ LV WKDW WKH FRUUHODWLRQ IXQFWLRQV RI WKH DJJUHJDWHG SURFHVVHV GR QRW GHFD\ WR DV P f§! RR 7KLV IHDWXUH LV LQ FRQWUDVW WR WUDGLWLRQDO PRGHOV DOO RI ZKLFK KDYH WKH SURSHUW\ WKDW WKH FRUUHODWLRQ IXQFWLRQ RI WKHLU DJJUHJDWHG SURFHVVHV GHFOLQH DV P f§} RR LH UPfUf DV L f§! RR IRU U f 7ZR RI WKH PRVW ZHOONQRZQ VHOIVLPLODU SURFHVVHV DUH IUDFWLRQDO *DXVVLDQ QRLVH I*Qf DQG LWV LQFUHPHQWDO YHUVLRQ IUDFWLRQDO %URZQLDQ PRWLRQ I%Pf ZKLFK ZLOO EH GLVFXVVHG LQ GHWDLO ODWHU LQ WKLV FKDSWHU

PAGE 44

6HOIVLPLODULW\ DQG /RQJ5DQJH 'HSHQGHQFH $ SURFHVV VDWLVI\LQJ f LV VDLG WR H[KLELW ORQJUDQJH GHSHQGHQFH /5'f 6XFK SURFHVVHV DUH FKDUDFWHUL]HG E\ DQ DXWRFRUUHODWLRQ IXQFWLRQ UUf ZKLFK GHFD\V K\SHUEROLFDOO\ DV U LQFUHDVHV 7KLV LPSOLFDWLRQ VKRZV WKDW UUf LV QRQVXPPDEOH LH 6Rr UUf ZKLFK FRQWUDVWV ZLWK PDQ\ FRQYHQWLRQDO VKRUWUDQJH GHSHQGHQW 65'f SURFHVVHV VXFK DV 3RLVVRQEDVHG SURFHVVHV ZKLFK KDYH DQ DXWRFRUUHODWLRQ IXQFWLRQ UUf ZKLFK GHFUHDVHV H[SRQHQWLDOO\ DV U LQFUHDVHV DQG LV FRQVHTXHQWO\ VXPPDEOH QDPHO\ ARr U7f rr )XUWKHUPRUH WKH QRQVXPPDELOLW\ RI WKH FRUUHODWLRQV FDSn WXUHV WKH LQWXLWLRQ EHKLQG /5' QDPHO\ WKDW ZKLOH LQGLYLGXDO YDOXHV RI UUf DUH VPDOO IRU ODUJH U WKHLU FXPXODWLYH HIIHFW LV RI LPSRUWDQFH DQG SURGXFHV IHDWXUHV ZKLFK DUH GUDVWLFDOO\ GLIIHUHQW IURP 65' SURFHVVHV 3LnRSRVLWLRQ $Q DV\PSWRWLFDOO\ VHOIVLPLODU SURFHVV ;Wf ZLWK +XUVW SDUDPHWHU + H[KLELWV /5' DV + DQG 65' DV + 3URRI )URP f WKH DXWRFRUUHODWLRQ IXQFWLRQ RI ;A JLYHQ +XUVW SDUDPHWHU + LV UUf ?>^W Of+ W f§ Of+ f§ W+@ :H KDYH UUf LUm>Uf§fm W f§f W@ U U LnULLa^U>L UBfLI f§ Waf+ f§ @` f f+ +U +>+ fU f f f f DQG UBf+ +Wa ++ fUa f f f f &RPELQLQJ f f DQG f ZH REWDLQ

PAGE 45

)LJXUH $XWRFRUUHODWLRQ IXQFWLRQ DQG ORQJUDQJH GHSHQGHQFH UUf sWKa^++ f Uf@ f ZKHUH Uf f§ D?7a DU f f f 6LQFH Uf DV U f§! RR ZH KDYH UU +f +^+ fU+a f )URP f LW LV REYLRXV WKDW UUf LV VXPPDEOH LI + DQG QRQVXPPDEOH RWKHUZLVH ’ )LJXUH GHSLFWV UUf IRU VHYHUDO YDOXHV RI + 2I SDUWLFXODU QRWLFH LV U Uf GHFD\V WR DV W f§! RR PXFK VORZHU ZKHQ + WKDQ ZKHQ + $V D UHVXOW UUf WHQGV WR VR VORZO\ WKDW ARr UUf rrL ZKLFK OHDGV WR WKH /5' RI ;Wf

PAGE 46

6HOIVLPLODULW\ DQG +XUVW (IIHFW 6HOIVLPLODU SURFHVVHV DOVR SURYLGH DQ HOHJDQW H[SODQDWLRQ RI DQ HPSLULFDO ODZ NQRZQ DV +XUVWfV ODZ RU WKH +XUVW HIIHFW )RU D JLYHQ VHW RI REVHUYDWLRQV $G $ ;Q ZLWK VDPSOH PHDQ ;Qf DQG VDPSOH YDULDQFH 6Qf WKH fUHVFDOHG DGMXVWHG UDQJHf RU 56 VWDWLVWLFf LV JLYHQ E\ PD[ GLG GQf PLQ G[ G GQf 6^Qf ZKHUH GMW ;L ; ;N f§ N ‘ ;Qf N Q f +XUVW >@ IRXQG WKDW PDQ\ QDWXUDOO\ RFFXUULQJ WLPH VHULHV ZHOO UHSUHVHQWHG E\ WKH UHODWLRQ (>5Qf6Qf@ f§! D f Q+ DV Q f§r‘ f ZLWK +XUVW SDUDPHWHU + QRUPDOO\ DURXQG DQG D D ILQLWH SRVLWLYH FRQVWDQW LQn GHSHQGHQW RI Q +RZHYHU LI WKH REVHUYDWLRQV FRPH IURP D VKRUWUDQJH GHSHQGHQW 65'f SURFHVV WKHQ LW KDV EHHQ VKRZQ LQ >@ WKDW (>5Qf6Qf@ f§! E ‘ Qrn DV Q f§! f ZLWK E DQRWKHU ILQLWH SRVLWLYH FRQVWDQW LQGHSHQGHQW RI Q $ 0DWODE VFULSW IRU +XUVW SDUDPHWHU HVWLPDWLRQ EDVHG RQ 56 VWDWLVWLF LV LQFOXGHG LQ $SSHQGL[ $O 6HOIVLPLODULW\ DQG 6ORZO\ 'HFD\LQJ 9DULDQFHV $QRWKHU LPSRUWDQW IHDWXUH RI VHOIVLPLODU SURFHVVHV LV WKDW WKH YDULDQFH RI WKH DULWKPHWLF PHDQ ID GHFUHDVHV VORZHU WKDQ WKH UHFLSURFDO RI WKH VDPSOH VL]H P 5HVHDUFK >@ VKRZV WKDW f LV HTXLYDOHQW WR YDU;Af f§! DPa3 DV P f f 5Qf6Qf

PAGE 47

ZKHUH D LV D ILQLWH SRVLWLYH FRQVWDQW LQGHSHQGHQW RI P DQG f§ + 2Q WKH RWKHU KDQG IRU VKRUWUDQJH GHSHQGHQW SURFHVVHV IRU ZKLFK $nA Nf DV P f§} 22 LW FDQ EH VKRZQ DV ZHOO WKDW YDU;nPnff f§! EPaO DV P f§} RR f ZKHUH E LV D ILQLWH SRVLWLYH FRQVWDQW LQGHSHQGHQW RI P 7KH +XUVW SDUDPHWHU HVWLPDn WLRQ EDVHG RQ f LV FDOOHG 9DULDQFHWLPH SORW YWSf $ 0DWODE VFULSW IRU YWS LV LQFOXGHG LQ $SSHQGL[ $ 0RGHOLQJ RI 6HOI6LPLODULW\ )RU GHFDGHV VHOIVLPLODULW\ KDV EHHQ D SKHQRPHQRQ WKDW KDV EHHQ VWXGLHG H[n WHQVLYHO\ LQ VHYHUDO RWKHU ILHOGV LQFOXGLQJ K\GURORJ\ >@ DQG HFRQRPLFV >@ %HFDXVH RI WKLV H[KDXVWLYH UHVHDUFK VHYHUDO IRUPDO PDWKHPDWLFDO PRGHOV KDYH DOUHDG\ EHHQ GHYHORSHG ZKLFK H[KLELW VHOIVLPLODULW\ +RZHYHU GXH WR WKH DEVWUDFW QDWXUH RI WKHVH PRGHOV D SUDFWLFDO LQWHUSUHWDWLRQ IRU WKH VHOIVLPLODULW\ LV GLIILFXOW WR ILQG 7ZR VXFK PRGHOV DUH SUHVHQWHG LQ WKLV VHFWLRQ WKH IUDFWLRQDO *DXVVLDQ QRLVH I*Qf DQG WKH IUDFWLRQDO DXWRUHJUHVVLYH LQWHJUDWHG PRYLQJ DYHUDJH PRGHOV I$5,0$f :H DOVR SUHn VHQWHG IUDFWLRQ %URZQLDQ PRWLRQ I%Pf WKH FRPSDQLRQ RI I*Q DQG D FRQVWUXFWLRQ EDVHG RQ DJJUHJDWLQJ PDQ\ VLPSOH UHQHZDO UHZDUGV SURFHVVHV 7KH ODWWHU LV RI SDUn WLFXODU VLJQLILFDQFH WR WKLV UHVHDUFK DV LW SURYLGHV D SUDFWLFDO LQWHUSUHWDWLRQ IRU WKH VHOIVLPLODULW\ IRXQG LQ (WKHUQHW /$1 WUDIILF )UDFWLRQDO %URZQLDQ 0RWLRQ 7KH RUGLQDU\ %URZQLDQ PRWLRQ %Wf GHVFULEHV WKH PRYHPHQW RI D SDUWLFOH LQ D OLTXLG VXEMHFWHG WR FROOLVLRQ DQG RWKHU IRUFHV 7KLV PRWLRQ LV D VHOIVLPLODU SURFHVV ZLWK +XUVW SDUDPHWHU DQG LQGHSHQGHQW *DXVVLDQ LQFUHPHQWV VXFK WKDW %Wf _W_%f f

PAGE 48

DQG (>%W[ Wf %WLffrLf@ f )URP f DQG f LW IROORZV WKDW (^>%W[ Uf %Af`` _U_ f (>%^ f@ f 0DQGHOEURW >@ GHILQHV IUDFWLRQDO %URZQLDQ PRWLRQ I%Pf ZLWK +QUVW SDUDPn HWHU + DV EHLQJ WKH PRYLQJ DYHUDJH RI G%Wf LQ ZKLFK SDVW LQFUHPHQWV RI %Wf DUH ZHLJKWHG E\ WKH NHUQHO W f§ Vf+an PUZf Vf MIr f ZKHUH 7 GHQRWHV WKH JDPPD IXQFWLRQ 7U f UUUf DQG 7N f N? DV N EHLQJ DQ LQWHJHU %\ GHILQLWLRQ LQ f ZH KDYH %+Wf W+%+ f f 6HYHUDO LPSRUWDQW SURSHUWLHV RI I%P DUH DV IROORZV %Kf (^%+Wf` L Uf %+W[ff` ?W?+ ‘ (^%+^ f` (^%+^W[f%+Wf` ??W[?+ ?K?+ ?W[ W?+f ‘ (^%K f` )UDFWLRQDO *DXVVLDQ 1RLVH 7KH LQFUHPHQWV RI I%P *Qf IRUP D VWDWLRQDU\ VHTXHQFH FDOOHG IUDFWLRQDO *DXVVLDQ QRLVH I*Qf

PAGE 49

*Qf %+Q f %+Qf f $ IUDFWLRQDO *DXVVLDQ QRLVH I*Qf LV D VWDWLRQDU\ *DXVVLDQ SURFHVV ZLWK PHDQ cMO YDULDQFH FU DQG DXWRFRUUHODWLRQ IWPFWLRQ UUf Of_U ?+ ?W?+ ?W ?+f f 3URP 3URSRVLWLRQ ZH SURYH WKDW UUf f§! ++ f§ f?W?+a DV W RR ZKLFK PHDQV I*Q H[KLELWV /5' ,W FDQ DOVR EH VKRZQ WKDW WKH DJJUHJDWHG SURFHVVHV ;A DOO KDYH WKH VDPH GLVWULEXWLRQ DV ; IRU DOO + 7KXV E\ f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n IRUPHG LQWR D VDPSOH SDWK ZLWK DUELWUDU\ PDUJLQDO GLVWULEXWLRQ ZKLOH PDLQWDLQLQJ WKH +XUVW SDUDPHWHU + >@ )UDFWLRQDO $5,0$SGTf 3URFHVVHV $XWRUHJUHVVLYHPRYLQJ DYHUDJH $50$f SURFHVVHV DUH RIWHQ XVHG IRU PRGHOLQJ HPSLULFDO WLPH VHULHV /HW S DQG T EH QRQQHJDWLYH LQWHJHUV WKHQ WKH VHTXHQFH ^;Q` LV FDOOHG $50$STf LI LW VDWLVILHV WKH HTXDWLRQV S 4 ;Q a A @ 4N;Qf§N 7 f @ ENAQf§NL f IF IF ZKHUH ^ZN N ` DUH LLG UDQGRP YDULDEOHV DQG DUH DVVXPHG WR EH HLWKHU *DXVVLDQ RU QRQ*DXVVLDQ 7KH ^DA` DUH UHIHUUHG WR DV f$5 SDUDPHWHUVf ZKLOH ^EA` DUH FDOOHG f0$ SDUDPHWHUVf )RU FRQYHQLHQFH GHQRWH

PAGE 50

e!ff f§ D?' f§ f§ DS'S f DQG 'f E EL' ET'? f ZKHUH LV D GHOD\ RSHUDWRU LH 'N;Q ;QN ,W IROORZV WKDW f FDQ EH UHZULWWHQ DV !^'f;Q 4'fZQ f RU ;Q a'f4'fZQ f :H QRZ WXUQ WR IUDFWLRQDO $5,0$ WLPH VHULHV /HW $ EH WKH GLIIHUHQFH RSHUDn WRU GHILQHG E\ $;Q ;Q f§ $7QBL f§ 'f;Q $ IUDFWLRQDO $5,0$SGTf SURFHVV LV GHILQHG DV D VWRFKDVWLF SURFHVV ; ;N N f UHSUHVHQWHG E\ ^'f$G;Q 4'fZN f 1RWH WKDW WKH SDUDPHWHU G LQ $G LV DOORZHG WR WDNH IUDFWLRQDO YDOXHV HLWKHU SRVLWLYH RU QHJDWLYH /RJLFDOO\ WKH IUDFWLRQDO GLIIHUHQFH RSHUDWRU IROORZV DV $G 'fG <&GNf^'fN f N ZKHUH N E f§ Q f§ FGNf Q 7N Gf f M UGfUIF fn )RU H[DPSOH LI S T DQG G LV D QRQQHJDWLYH LQWHJHU WKHQ $G;Q ZQ GHVFULEHV D PRGHO ZKHUH ;Q GLIIHUHQFHG G WLPHV \LHOGV D VHTXHQFH RI LLG UDQGRP YDULDEOHV ZQ $V D FRQVHTXHQFH $5,0$f FKDUDFWHUL]HV D UDQGRP VHTXHQFH JHQHUDWHG E\ UDQGRP ZDON 2I SDUWLFXODU QRWLFH LV LW KDV EHHQ VKRZQ LQ >@ WKDW IRU G H f

PAGE 51

D IUDFWLRQDO $5,0$S G Tf SURFHVV VDWLVILHV HTXDWLRQV (T ZKLFK PDNHV LW DQ DV\PSWRWLFDO VHOIVLPLODU SURFHVVHV ZLWK /5' DQG +XUVW SDUDPHWHU + f§ G )UDFWLRQDO $5,0$S G Tf SURFHVVHV ZHUH ILUVW LQWURGXFHG LQ >@ 2QH DGYDQn WDJH RI IUDFWLRQDO $5,0$S G Tf SURFHVVHV RYHU IUDFWLRQDO *DXVVLDQ QRLVH LV WKHLU IOH[LELOLW\ ZLWK UHVSHFW WR PRGHOLQJ ERWK VKRUWUDQJH DQG ORQJUDQJH GHSHQGHQFH PDNLQJ WKHP EHWWHU VXLWHG WR PRGHOLQJ SKHQRPHQRQ VXFK DV 9%5 YLGHR WUDIILF >@ 6HOI6LPLODULW\ 7KURXJK $JJUHJDWLRQ &RQVLGHU D QXPEHU RI LQGHSHQGHQW VRXUFHV ZKLFK DOWHUQDWH EHWZHHQ WKH VWDWHV fRQf DQG fRIIf UHQHZDOUHZDUGV SURFHVVHVf ZKHUH WKH DPRXQW RI WLPH VSHQW LQ HDFK VWDWH LV UDQGRPO\ GLVWULEXWHG ZLWK D KHDY\WDLOHG GLVWULEXWLRQ $Q H[DPSOH RI VXFK D KHDY\WDLOHG GLVWULEXWLRQ LV WKH VWDEOH 3DUHWR GLVWULEXWLRQ VHH $SSHQGL[ % RI >@f ZLWK SDUDPHWHU D ,I ZH FRQVWUXFW D QHZ VHTXHQFH FRQVLVWLQJ RI WKH QXPEHU RI VRXUFHV ZKLFK DUH fRQf DW SRLQWV LQ WLPH VHSDUDWHG E\ D IL[HG OHQJWK LQWHUYDO WKHQ LW KDV EHHQ VKRZQ LQ >@ WKDW WKH UHVXOWLQJ VHTXHQFH ZLOO EH DV\PSWRWLFDOO\ VHOIVLPLODU 7KH QRWLRQ RI JHQHUDWLQJ D VHOIVLPLODU SURFHVV E\ DJJUHJDWLQJ D QXPEHU RI VLPSOH UHQHZDO UHZDUGV SURFHVVHV SURYLGHV LQWXLWLRQ LQWR WKH VHOIVLPLODULW\ RI (WKHUQHW WUDIILF DV ZHOO DV LQWR WKH QDWXUH RI D VLQJOH VRXUFH RQ DQ (WKHUQHW QHWZRUN ,I ZH FRQVLGHU HDFK ZRUNVWDWLRQ RU PRUH OLNHO\ HDFK XVHUf DWWDFKHG WR DQ (WKHUQHW QHWZRUN DV D UHQHZDOUHZDUGV SURFHVV ZKHUH WKH SURFHVV LV fRQf LI WKH ZRUNVWDWLRQ LV FRPPXQLFDWLQJ RYHU WKH QHWZRUN DQG fRIIf RWKHUZLVH WKHQ ZH VHH WKDW WKH DPRXQW RI WUDIILF RQ WKH QHWZRUN FDQ EH WKRXJKW RI DV WKH DJJUHJDWLRQ RI D QXPEHU RI UHQHZDO UHZDUGV SURFHVVHV 7KLV DOVR OHDGV WR WKH FRQFOXVLRQ WKDW WKH FRPPXQLFDWLRQ EXUVWV SURGXFHG E\ LQGLYLGXDO XVHUV KDYH D KHDY\WDLOHG GLVWULEXWLRQ 7KLV FRQFOXVLRQ LV VXSSRUWHG E\ LQ GHSWK DQDO\VLV RI (WKHUQHW >@ DQG ZLGH DUHD QHWZRUN WUDIILF WUDFHV >@

PAGE 52

6HOIVLPLODU 7UDIILF 7UDFH *HQHUDWRUV :KLOH PRVW QHWZRUN PDQDJHPHQW WRROV VXFK DV +3fV 2SHQ9LHZ DQG ,%0fV 1HW 9LHZ KDYH EXLOWLQ FDSDELOLWLHV WR PRQLWRU DQG JDWKHU WUDFH GDWD QRW HYHU\RQH KDV DFFHVV WR VXFK WRROV 2QH RI WKH PRVW SRSXODU PRQLWRULQJ WRROV GHVLJQHG DV XVHU FRPPDQGV LV fWFSGXPSf ZKLFK LV XVHIXO WR JHW DQ RYHUYLHZ RI ZKDW W\SH RI RU KRZ PDQ\ SDFNHWV DUH RQ D QHWZRUN IRU D JLYHQ WLPH SHULRG 7KH SUREOHP LV WKDW QRW PDQ\ SHRSOH IHHO FRPIRUWDEOH UHDGLQJ DQG UHFRUGLQJ WKHP PRVWO\ GXH WR WKH FRQFHUQV RI LQGLYLGXDO SULYDF\ RU QHWZRUN VHFXULW\ +RZHYHU WR GR VLPXODWLRQ VWXGLHV D ORW RI WUDFH GDWD LV QHHGHG DQG JDWKHULQJ DQG VWRULQJ WKLV GDWD QRW RQO\ LV WLPH FRQVXPLQJ EXW DOVR UHTXLUHV KXJH VWRUDJH VSDFH :KDW LV QHHGHG LV D IDVW PHWKRG RI JHQHUDWLQJ V\QWKHWLF WUDIILF WUDFHV IRU XVH LQ VLPXODWLRQV &XUUHQWO\ D QXPEHU RI DSSURDFKHV WR JHQHUDWH VHOIVLPLODU VDPSOH SDWKV DUH DYDLODEOH DQG ZH LQWURGXFH VHYHUDO SRSXODU RQHV LQ WKH IROORZLQJ VHFWLRQV 5DQGRP 0LGSRLQW 'LVSODFHPHQW $OJRULWKP 7KH PRVW ZHOONQRZQ DOJRULWKP IRU JHQHUDWLQJ I%P LV WKH f5DQGRP 0LGn SRLQW 'LVSODFHPHQWf 50'f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

PAGE 53

)LJXUH )UDFWLRQDO %URZQLDQ PRWLRQ DQG IUDFWLRQDO *DXVVLDQ QRLVH

PAGE 54

$JJUHJDWLRQ RI 5HQHZDO 3URFHVVHV $ VHOIVLPLODU WUDFH JHQHUDWRU GHVFULEHG LQ >@ FRQVLVWV RI DJJUHJDWLQJ D QXPn EHU RI UHQHZDO UHZDUGV SURFHVVHV WR REWDLQ D VHOIVLPLODU SURFHVV 7KLV DSSURDFK KDV WKH SUREOHP WKDW RQH PXVW WUDGH RII VSHHG IRU WKH GHJUHH RI VHOIVLPLODULW\ DV LQn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f UXQQLQJ WLPH WR JHQHUDWH D VDPSOH SDWK RI OHQJWK Q 7KH DXWKRUV RI >@ UHSRUW WDNLQJ &38 KRXUV WR JHQHUDWH SRLQWV XVLQJ WKLV PHWKRG

PAGE 55

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n DQWHHG WR KDYH WKH DXWRFRUUHODWLRQ VWUXFWXUH DQG KHQFH VHOIVLPLODULW\f RI IUDFWLRQDO *DXVVLDQ QRLVH 7KLV PHWKRG KDV WKH SURSHUW\ WKDW LW LV IDVW JHQHUDWLQJ SRLQWV WDNHV VHFRQGV RQ D 63$5&VWDWLRQ ,3;f DQG WKDW LW GRHV QRW VXIIHU IURP WKH ELDVHV DVVRFLDWHG ZLWK WKH 5DQGRP 0LGSRLQW 'LVSODFHPHQW DOJRULWKP %HFDXVH LW LV EDVHG RQ WKH ))7 WKLV PHWKRG LV UHIHUUHG WR E\ LWV DXWKRUV DV WKH ))7 PHWKRG RI JHQHUDWLQJ VHOIVLPLODU VDPSOH SDWKV $ 0DWODE VFULSW EDVHG RQ WKLV PHWKRG LV LQFOXGHG LQ $SSHQGL[ %

PAGE 56

&+$37(5 &21*(67,21 $1' )/2: &21752/ 7KH %,6'1 ZKLFK LV EDVHG RQ WKH $70 WHFKQLTXH LV GHVLJQHG WR WUDQVSRUW D ZLGH YDULHW\ RI WUDIILF FODVVHV VDWLVI\LQJ D UDQJH RI WUDQVIHU FDSDFLW\ QHHGV DQG QHWZRUN SHUIRUPDQFH REMHFWLYHV 7KH XQGHUDFKLHYHPHQW LI DQ\ LV XVXDOO\ FDXVHG E\ QHWZRUN FRQJHVWLRQ ZKLFK FRXOG KDSSHQ ZKHQHYHU WKH RYHUDOO LQVWDQWDQHRXV LQSXW UDWH LV JUHDWHU WKDQ WKH OLQN FDSDFLW\ IRU D FHUWDLQ SHULRG RI WLPH LH e$ ,!& f ZKHUH $ UHSUHVHQWV WKH LQVWDQWDQHRXV LQSXW UDWH RI DQ DFWLYH VRXUFH DQG & WKH OLQN FDSDFLW\ 7R EH PRUH JHQHUDO FRQJHVWLRQ LQ $70 QHWZRUNV FDQ EH FDXVHG E\ VWDWLVWLFDO IOXFWXDWLRQ RI WUDIILF IORZV RU QHWZRUN IDLOXUHV >@ ,Q OLJKW RI WKH GLYHUVH WUDIILF FKDUDFWHULVWLFV LQ $70 QHWZRUNV WKH GHVLJQ RI D VFKHPH WR DGHTXDWHO\ FRQWURO WKH GDWD IORZ LQWR $70 QHWZRUNV VR DV WR DYRLG FRQJHVWLRQ EHFRPHV D FULWLFDO IDFWRU IRU WKH VXFFHVV RI %,6'1 $VLGH IURP WKH SULPDU\ UROH RI FRQJHVWLRQ DQG IORZ FRQWURO VFKHPHV WR SURWHFW WKH QHWZRUN DQG WKH XVHU LQ RUGHU WR DFKLHYH QHWZRUN SHUIRUPDQFH REMHFWLYHV DQ DGGLWLRQDO UROH LV WR RSWLPL]H WKH XVH RI WKH QHWZRUN UHVRXUFHV VR DV WR DFKLHYH EHWWHU XWLOL]DWLRQ HIILFLHQF\ *HQHUDO )UDPHZRUN IRU &RQJHVWLRQ 0DQDJHPHQW DQG &RQWURO )RU D FRQJHVWLRQ PDQDJHPHQW DQG FRQWURO VFKHPH WR EH DEOH WR VXSSRUW D VHW RI 4R6 FODVVHV VXIILFLHQW IRU DOO IRUHVHHDEOH VHUYLFHV DQG PDLQWDLQ PLQLPXP QHWZRUN DQG HQGV\VWHP FRPSOH[LW\ ZKLOH PD[LPL]H QHWZRUN XWLOL]DWLRQ WKH IROORZLQJ IXQFWLRQV

PAGE 57

IRUP D IUDPHZRUN IRU PDQDJLQJ DQG FRQWUROOLQJ FRQJHVWLRQ LQ $70 QHWZRUNV DQG PD\ EH XVHG LQ DSSURSULDWH FRPELQDWLRQV >@ f 5HVRXUFH 0DQDJHPHQW 50f > @ 50 SURYLGHV IRU DOORFDWLQJ QHWZRUN UHVRXUFHV LQ RUGHU WR VHSDUDWH WUDIILF IORZ DFFRUGLQJ WR WKH 4R6 RQ WUDIILF FRQn WUDFW 7ZR FULWLFDO UHVRXUFHV LQ $70 QHWZRUNV DUH EXIIHU VSDFH DQG WUXQN EDQGZLGWK f &RQQHFWLRQ RU &DOOf $GPLVVLRQ &RQWURO &$&f > @ &$& GHILQHV D VHW RI DFWLRQ WDNHQ E\ WKH QHWZRUN GXULQJ FDOO VHWXS RU UHQHJRWLDWLQJ SKDVH LQ RUGHU WR GHWHUPLQH ZKHWKHU D 9&& RU 93& UHTXHVW FDQ EH JUDQWHG RU QRW $SSDUHQWO\ URXWLQJ LV SDUW RI &$& DFWLRQV 2QH LPSRUWDQW FODVV RI DSSOLFDWLRQV RI $70 QHWZRUNV LV LQWHUDFWLYH SURJUDPV VXFK DV YLGHRFRQIHUHQFLQJ ,Q RUGHU WR SURYLGH 4R6 UHTXLUHPHQWV IRU DOO FRQQHFWLRQV &$& LV DQ HVVHQWLDO PHFKDQLVP f 7UDIILF 6KDSLQJ 76f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f DQG %XIIHULQJ 3&55 FDQ EH DFKLHYHG E\ RSHUDWLQJ WKH VHQGLQJ WHUPLQDO DW D SHDN UDWH OHVV WKDQ WKDW LQ WKH WUDIILF FRQWUDFW UHGXFLQJ WKH SRVVLELOLW\ RI FRQIRUPDQFH YLRODWLRQ %XIIHULQJ RSHUDWHV LQ FRPELQDWLRQ ZLWK WKH OHDN\ EXFNHW DOJRULWKP WR HQVXUH WKDW FHOOV ZLOO QRW YLRODWH WKH WUDIILF SDUDPHWHUV RI WKH FRQWUDFW E\ EXIIHULQJ FHOOV XQWLO OHDN\ EXFNHW ZRXOG DGPLW WKHP f 3ULRULW\ &RQWURO >@ %\ XVLQJ &HOO /RVV 3ULRULW\ ELW WKH XVHU PD\ JHQHUDWH WUDIILF IORZV ZLWK GLIIHUHQW SULRULWLHV $Q $70 VZLWFK PD\ VHOHFWLYHO\ GLVFDUG

PAGE 58

FHOOV ZLWK ORZ SULRULW\ LI QHFHVVDU\ WR SURWHFW WKH 4R6 RI FHOOV ZLWK KLJK SULRULW\ DV PXFK DV SRVVLEOH f 8VDJH 3DUDPHWHU &RQWURO 83&f >@ LV GHILQHG DV WKH VHW RI DFWLRQV WDNHQ E\ WKH QHWZRUN WR PRQLWRU DQG FRQWURO WUDIILF LQ WHUPV RI WUDIILF RIIHUHG DQG IDLUQHVV RI $70 FRQQHFWLRQV 7KH PDLQ SXUSRVH LV WR SURWHFW QHWZRUN UHVRXUFH IURP PDOLFLRXV RU XQLQWHQWLRQDO PLVEHKDYLRUV ZKLFK FDQ DIIHFW WKH 4R6 RI RWKHU DOn UHDG\ HVWDEOLVKHG FRQQHFWLRQV E\ GHWHFWLQJ YLRODWLRQV RI QHJRWLDWHG SDUDPHWHUV DQG WDNLQJ DFWLRQV DFFRUGLQJO\ f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fWUDIILF VKDSLQJ DOJRULWKPVf DUH RSHQ ORRS LQ WKH VHQVH WKDW WKH SDUDPHWHU FDQQRW EH FKDQJHG G\QDPLFDOO\ LI FRQJHVWLRQ LV GHWHFWHG DIWHU FRQJHVWLRQ 2Q WKH FRQWUDU\ LQ D FORVHG ORRS VFKHPH VRXUFHV DUH LQIRUPHG G\QDPLFDOO\ DERXW WKH FRQJHVWLRQ VWDWH RU UHVRXUFH XVDJH RI WKH QHWZRUN DQG DUH DVNHG WR DGMXVWHG WKHLU LQSXW UDWH DFFRUGLQJO\ ,Q JHQHUDO WKLV LQIRUPDWLRQ LV FDUULHG E\ DFNQRZOHGJPHQW RU 50 FHOOV YLD KRSE\KRS

PAGE 59

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n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f DQG ZLQGRZ IORZ FRQWURO (DFK OLQN FRQVLVWV RI D VHQGHU QRGH DQG D UHFHLYHU QRGH (DFK QRGH PDLQWDLQV D VHSDUDWH TXHXH IRU HDFK 9& 7KH UHFHLYHU PRQLWRUV TXHXH OHQJWK RI HDFK 9& DQG GHWHUPLQHV WKH QXPEHU RI FHOOV WKDW WKH VHQGHU FDQ WUDQVPLW RQ WKDW 9& 7KLV QXPEHU LV FDOOHG fFUHGLWf 7KH VHQGHU WUDQVPLWV RQO\ DV PDQ\ FHOOV DV DOORZHG E\ WKH FUHGLW ,Q WKH FDVH RI D VLQJOH DFWLYH 9& WKH FUHGLW PXVW EH ODUJH HQRXJK WR DOORZ WKH ZKROH OLQN WR EH IXOO DW DOO WLPHV ,Q RWKHU ZRUGV &UHGLW /LQN &HOO 5DWH [ /LQN 5RXQG 7ULS 'HOD\ f ZKHUH RQH FDQ FRPSXWH WKH OLQN FHOO UDWH E\ GLYLGLQJ WKH OLQN EDQGZLGWK LQ 0ESV E\ WKH FHOO VL]H LQ ELWV

PAGE 60

7KH FUHGLWEDVHG VFKHPH DV GHVFULEHG VR IDU LV DOVR UHIHUUHG WR DV f)ORZ &RQn WUROOHG 9LUWXDO &LUFXLW )&9&ff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f LQ WKH QH[W SHULRG 7KH DGDSWLYH )&9& UHGXFHV WKH EXIIHU UHTXLUHPHQWV FRQVLGHUDEO\ EXW DOVR WKH IXOO FDSDFLW\ RI WKH OLQN HYHQ LI WKHUH DUH QR RWKHU XVHUV 5DWH%DVHG 6FKHPH 7KH RULJLQDO SURSRVDO IRU D UDWHEDVHG DSSURDFK FRQVLVWV RI HQGWRHQG FRQWURO XVLQJ D VLQJOHELW WR FRQWURO WKH QHWZRUN ,Q WKH SURSRVDO WKH VZLWFKHV PRQLWRU WKHLU TXHXH OHQJWKV DQG LI FRQJHVWHG VHW WKH H[SOLFLW IRUZDUG FRQJHVWLRQ LQGLFDWLRQ ()&,f LQ WKH FHOO KHDGHUV 7KH GHVWLQDWLRQ PRQLWRUV WKHVH LQGLFDWLRQV IRU D SHULRGLF LQWHUYDO DQG VHQGV DQ 50 FHOO EDFN WR WKH VRXUFH 7KH VRXUFHV XVH DQ DGGLWLYH LQFUHDVH DQG PXOWLSOLFDWLYH GHFUHDVH DOJRULWKP WR DGMXVW WKHLU UDWHV 7KLV SDUWLFXODU DOJRULWKP

PAGE 61

XVHV D fQHJDWLYH IHHGEDFNf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f%DFNZDUG ([SOLFLW &RQJHVWLRQ 1RWLILFDWLRQf %(&1f VFKHPH 7KLV PHWKRG SUHn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

PAGE 62

,PSURSULHW\ RI 3RLVVRQ%DVHG 7UDIILF 0RGHOV 7KLV VHVVLRQ SUHVHQWV WKH VLPXODWLRQ HIIRUWV WR MXVWLI\ WKH SHUIRUPDQFH RI WKH IORZ FRQWURO VFKHPHV LQWURGXFHG LQ WKH SUHYLRXV VHFWLRQ E\ XVLQJ 3RLVVRQ SURFHVV WUDIILF PRGHOV DQG VHOIVLPLODU SURFHVV 663f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f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

PAGE 63

)LJXUH 1HWZRUN VFHQDULR IRU 3RLVVRQ YV 6HOIVLPLODU WUDIILF PRGHOV JRRGSXW LV H[SHFWHG IURP HDFK RI WKH HLJKW FRQQHFWLRQV 7KH PD[LPXP JRRGSXW WKDW PD\ EH DFKLHYHG LQ WKLV VFHQDULR LV 7KH +XUVW SDUDPHWHU RI WKH JHQHUDWHG VDPSOH SDWKV LV HVWLPDWHG E\ 9DULDQFHn WLPH SORW YWSf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

PAGE 64

)LJXUH 9DULDQFHWLPH SORW IRU V\QWKHVL]HG VDPSOH SDWKV YHULILHG E\ WKH UHVXOWV VKRZQ LQ WKLV )LJXUH >@ WKH $70 )RUXP KDV VHOHFWHG UDWH EDVHG IORZ FRQWURO VFKHPH WR EH XVHG IRU IORZ DQG FRQJHVWLRQ FRQWURO LQ $70 QHWZRUNV 7KXV LQ VWXG\LQJ WKH HIIHFWV RI VHOIVLPLODU WUDIILF RQ WKH QHWZRUN SHUIRUPDQFH ZH FRQFHQWUDWH RQ WKH %(&1 VFKHPH ZLWK RU ZLWKRXW WKH VORZ VWDUW RSWLRQ 1RWH WKDW LQ WKLV VHVVLRQ fUDWHEDVHG VFKHPH ,f DQG fUDWHEDVHG VFKHPH ,,f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

PAGE 65

7UDIILF PRGHO UDWHEDVHG $b UDWHEDVHG $b 3RLVVRQ 663 + b b 663 + b b 7DEOH 3HUIRUPDQFH FRPSDULVRQ RI 3RLVVRQ SURFHVV DQG 663 WUDIILF PRGHOV 4 i VLPXODWLRQ WLPH VHFf )LJXUH 3HUIRUPDQFH FRPSDULVRQ RI GLIIHUHQW IORZ FRQWURO VFKHPHV PRGHO IRU 6HOIVLPLODU WUDIILF ZLWK + UHQGHUV b DQG b RI RYHUHVWLPDWLRQV RI WKH JRRGSXW IRU WKH %(&1 VFKHPH ZLWKRXW DQG ZLWK VORZVWDUW UHVSHFWLYHO\ :KLOH XVLQJ D 3RLVVRQ SURFHVV PRGHO IRU 6HOIVLPLODU WUDIILF ZLWK + UHVXOWV LQ b DQG b RI RYHUHVWLPDWLRQV RI WKH JRRGSXW IRU WKH %(&1 VFKHPH ZLWKRXW DQG ZLWK VORZVWDUW UHVSHFWLYHO\ 7KHVH UHVXOWV GHPRQVWUDWH WKH IDLOXUH RI 3RLVVRQ SURFHVVHV DV WUDIILF PRGHOV IRU WUDIILF H[KLELWLQJ VHOIVLPLODULW\ &ODVVLFDO &RQWURO 7KHRU\ DQG &RQJHVWLRQ &RQWURO )RU D V\VWHP DV VKRZQ LQ )LJXUH FODVVLFDO FRQWURO WKHRU\ LV 7LVXDOO\ YDOXDEOH DV D VRXUFH RI LQVLJKW LQ WKH GHVLJQ RI IHHGEDFN FRQWUROOHUV ,Q D IHHGEDFN IORZ FRQWURO

PAGE 66

2 VLPXODWLRQ WLPH VHFf )LJXUH 3HUIRUPDQFH FRPSDULVRQ RI UDWHEDVHG VFKHPH )LJXUH 3HUIRUPDQFH FRPSDULVRQ RI UDWHEDVHG VFKHPH ,,

PAGE 67

Er &RQWUROOHU 8QFHUWDLQW\ 1HWZRUN E $ } ? X )HHGEDFN )LJXUH *HQHULF )HHGEDFN &RQWURO 6\VWHP V\VWHP WKLV ILJXUH LV WKH W\SH RI V\VWHP XVHG WR HQVXUH WKDW WKH QHWZRUN WKH FRQn WUROOHG REMHFW FKDUDFWHUL]HG E\ LQSXW X LQWHUQDO VWDWHV [ RXWSXW \ DQG XQDYRLGDEOH XQFHUWDLQW\ VKRXOG SHUIRUP LQ D GHVLUHG PDQQHU 6HYHUDO LPSRUWDQW LQVLJKWV IURP FODVVLFDO FRQWURO WKHRU\ DUH DV IROORZV $ IHHGEDFN FRQWUROOHU PXVW FRPELQH GXDO IXQFWLRQ RI HVWLPDWLRQ DQG RI FRQWURO 7KH HVWLPDWLRQ UHGXFHV XQFHUWDLQW\ DERXW FXUUHQW VWDWH ZKRVH FXUUHQW YDOXHV DUH QHHGHG LQ RUGHU WR GHWHUPLQH FXUUHQW FRQWURO DFWLRQ )HHGEDFN LV WKH PHDQV E\ ZKLFK FRQWURO LV DFKLHYHG LQ WKH SUHVHQFH RI XQFHUn WDLQW\ 7KHUHIRUH DQ\ IHHGEDFN V\VWHP VKRXOG EHQHILW IURP D GHVLJQ DSSURDFK ZKLFK WDNH H[SOLFLW DFFRXQW RI XQFHUWDLQW\ 7ZR GLIIHUHQW W\SHV RI XQFHUWDLQW\ FDQ DIIHFW FRQWUROOHG REMHFWV XQFHUWDLQW\ DERXW FXUUHQW YDOXHV RI WKH LQWHUQDO VWDWHV DQG XQFHUWDLQW\ DERXW IXWXUH YDOn XHV UHVXOWLQJ IURP XQFHUWDLQW\ DERXW FRQWUROOHG REMHFW G\QDPLFV JRYHUQLQJ WKH HYROXWLRQ RI IXWXUH VWDWH 7KH FRQFHSW GUDZQ IURP WKH FODVVLFDO FRQWURO WKHRU\ ZLOO VHUYH DV D EDVLV IRU WKH SURSRVHG IORZ FRQWURO VFKHPHV LQ &KDSWHU

PAGE 68

&+$37(5 35(9(17,9( )/2: &21752/ 6&+(0( )25 6(/)6,0,/$5 75$)),& 5HFHQW PHDVXUHPHQW VWXGLHV KDYH VKRZQ WKDW WKH EXUVWLQHVV RI WKH WUDIILF LQ /$1V RU :$1V LV DVVRFLDWHG ZLWK ORQJUDQJH GHSHQGHQFH /5'f ,Q &KDSWHU ZH VKRZHG WKDW /5' FDQ EH HIILFLHQWO\ FKDUDFWHUL]HG E\ D TXHXHLQJ PRGHO ZLWK VHOIn VLPLODU DUULYDO SURFHVVHV VXFK DV IUDFWLRQDO *DXVVLDQ QRLVH I*Qf RU DXWRUHJUHVVLYH LQWHJUDWHG PRYLQJ DYHUDJH $5,0$f SURFHVVHV $V LQGLFDWHG LQ >@ WKH I*Q SURYLGHV WKH VLPSOHVW ZD\ WR PRGHO WKH /5' DQG VHOIVLPLODULW\ ,Q WKLV FKDSWHU ZH ZLOO FRQVHTXHQWO\ FRQVLGHU I*Q DV WKH DUULYDO SURFHVV LQ DQ $70 QHWZRUN HQYLURQPHQW 7R DYRLG WKH FRQJHVWLRQ DQG EXIIHU RYHUIORZ UHVXOWHG IURP EXUVW\ WUDIILF ZH SURSRVH D SUHYHQWLYH IORZ FRQWURO VFKHPH ZKLFK FRQVLVWV RI D QHZ FDOO DGPLVVLRQ FRQWURO &$&f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f ,Q DGGLWLRQ WKH DOJRULWKP SURYLGHV OHVV GHYLDWLRQ IURP WKH WDUJHW FHOO ORVV SUREDELOLW\ WKDQ RWKHU VWDWLVWLFDO DOORFDWLRQ VFKHPH VXFK DV f(TXLYDOHQW &DSDFLW\f

PAGE 69

,QWURGXFWLRQ 7KH 6HOIVLPLODU SURFHVV LV D UHDOLVWLF PRGHO IRU FKDUDFWHUL]LQJ WKH VWDWLVWLn FDO EHKDYLRUV RI WKH WUDIILF FRUUHVSRQGLQJ WR /$1V :$1V DQG 9%5 FRGHG YLGHR 6HOIVLPLODU SURFHVVHV DUH FKDUDFWHUL]HG E\ DQ K\SHUEROLF GHFD\ RI WKHLU DXWRFRUUHODn WLRQ IXQFWLRQ WKDW FDQQRW EH FDSWXUHG E\ WUDGLWLRQDO 3RLVVRQEDVHG SURFHVVHV 7KH EXUVWLQHVV RI PXOWLPHGLD WUDIILF UHSUHVHQWV WKH PDLQ UHDVRQ EHKLQG WKH GHYHORSPHQW RI VWDWLVWLFDO PXOWLSOH[LQJ VFKHPHV IRU WUDQVSRUW RI KHWHURJHQRXV WUDIILF RYHU EURDGEDQG QHWZRUNV VXFK DV $70 QHWZRUNV ,Q $70 QHWZRUNV FHOOV DUH VZLWFKHG DFFRUGLQJ WR LQIRUPDWLRQ FRQWDLQHG LQ WKHLU KHDGHUV EHWZHHQ $70 VZLWFKHV :KHQ FHOOV DUULYH DW D VZLWFK WKH\ PD\ EH VWRUHG LQ D EXIIHU DZDLWLQJ WUDQVPLVVLRQ WR WKHLU QHZ GHVn WLQDWLRQV ,I ZLWKLQ D FHUWDLQ SHULRG RI WLPH WKH QXPEHU RI DUULYLQJ FHOOV ZKLFK PD\ FRPH IURP GLIIHUHQW DQG RIWHQ EXUVW\ VRXUFHV LV ODUJHU WKDQ WKH QXPEHU RI FHOOV WKDW FDQ EH VHUYHG WKH EXIIHU PD\ RYHUIORZ DQG FHOOV PD\ EH ORVW ,Q YLGHR VHUYLFHV FHOO ORVV UHGXFHV SLFWXUH TXDOLW\ 7KH GHVLJQ RI IORZ DQG FRQJHVWLRQ FRQWURO LQ $70 QHWZRUNV LV WKHUHIRUH IDFHG ZLWK WKH FKDOOHQJH RI KRZ WR SURYLGH UHTXLUHG 4R6 DQG VWLOO PDLQWDLQ VXIILFLHQW QHWZRUN XWLOL]DWLRQ WR RSHUDWH DQ HFRQRPLFDO YLDEOH QHWZRUNV )ORZ DQG FRQJHVWLRQ FRQWURO SURFHGXUHV FDQ EH FODVVLILHG LQWR SUHYHQWLYH FRQn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f DQG EDQGZLGWK HQIRUFHPHQW 6LQFH $70 LV D FRQQHFWLRQRULHQWHG D

PAGE 70

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f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f DOVR FDOOHG QRQVWDWLVWLFDO DOORFDWLRQ LV WKH PRVW ZLGHO\ XVHG &$& GXH WR LWV VLPSOLFLW\ :KHQ D XVHU UHTXHVWV D FRQQHFWLRQ WKH GHFLVLRQ WR DFFHSW RU UHMHFW WKH UHTXHVW GHSHQGV RQ ZKHWKHU LWV SHDN FHOO UDWH 3&5f LV JUHDWHU RU OHVV WKDQ WKH DYDLODEOH UDWH 7KH PDMRU GUDZEDFN RI 35$ LV WKDW WKH QHWZRUN UHVRXUFHV ZLOO EH ZHOO XQGHUXWLOL]HG LI WKH UDWLR RI 3&5 WR 0&5 PHDQ FHOO UDWHf RI WKH FRQQHFWLRQ LV PXFK ODUJHU WKDQ RQH 8VLQJ 35$ IRU 9%5 VHUYLFHV LV RQH RI WKH EHVW H[DPSOHV RI WKLV VLWXDWLRQ *LYHQ D 3&50&5 HTXDO WR 5 WKH UHVRXUFH

PAGE 71

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f(TXLYDOHQW &DSDFLW\f >@ GHILQHG DV WKH VHUYLFH UDWH RI WKH TXHXH WKDW FRUUHVSRQGV WR D FHOO ORVV SUREDELOLW\ H %\ XVLQJ DQ LQWHUUXSWHG IOXLG SURFHVV DV WKH WUDIILF PRGHO >@ VKRZV WKDW JLYHQ D ILQLWH EXIIHU VL]H SHDN UDWH 5 0&5 S DQG WKH XWLOL]DWLRQ UDWH S WKH UHTXLUHG EDQGZLGWK WR DFKLHYH FHOO ORVV SUREDELOLW\ H LV f ZKHUH D f§OQHfSO f§ Sf5 &HOO /RVV 3UREDELOLW\ LQ 6HOIVLPLODU 4XHXLQJ 0RGHOV 7R HYDOXDWH WKH FHOO ORVV SUREDELOLW\ LQ DQ $70 FRQQHFWLRQ ZH VWXGLHG WKH SHUn IRUPDQFH RI D TXHXHLQJ V\VWHP ORDGHG E\ WKH FRPSRVLWLRQ RI LQGHSHQGHQW IUDFWLRQDO *DXVVLDQ QRLVH I*Qf SURFHVVHV :H GHVFULEHG D SURSRVHG GLIIXVLRQ DSSUR[LPDWLRQ IRU WKH QXPEHU RI DUULYDOV LQ WKH LQWHUYDO W@ FRQVLGHULQJ DQ KHDY\ WUDIILF UHQHZDO SURFHVVHV EHORZ 3URSRVLWLRQ ,Q D TXHXHLQJ V\VWHP ZLWK UHQHZDO DUULYDO SURFHVV ZLWK PHDQ S WKH FXPXODWLYH FRXQWV RI DUULYDO DWf FDQ EH DSSUR[LPDWHG E\ DWf SW \IS%Lf f

PAGE 72

ZKHUH %Wf LV D VWDQGDUG %URZQLDQ PRWLRQ QDPHO\ D *DXVVLDQ SURFHVV ZLWK LQGHSHQn GHQW LQFUHPHQWV f IROORZV IURP WKH HTXLYDOHQW SURSRVLWLRQ A rr!‘ ZKLFK FDQ EH SURYHG E\ &HQWUDO /LPLW 7KHRUHP >@ ([WHQGLQJ f WR WKH FDVH RI I*Q DUULYDO SURFHVVHV DV VXJJHVWHG LQ >@ DWf WKH QXPEHU RI DUULYDOV XS WR WLPH W FDQ EH UHSUHVHQWHG E\ DWf SW \S%+Wf f ZKHUH Y LV WKH YDULDQFH FRHIILFLHQW DQG %QWf LV D QRUPDOL]HG IUDFWLRQDO %URZQLDQ PRWLRQ I%Pf ZLWK ]HUR PHDQ DQG YDULDQFH ?W?+ /HW ^Qf GHQRWH WKH LQFUHPHQW RI $Wf LQ WKH LQWHUYDO Q f§ f7Q7? :H REWDLQ Qf DQ7f f§ DQ f§ f7f S7 \-Z^%+Q7f f§ %+Q f§ f7f` S7 Y@*Q7f f ZKHUH *Qf LV D IUDFWLRQDO *DXVVLDQ QRLVH ZLWK ]HUR PHDQ DQG D YDULDQFH HTXDO WR RQH )URP f LW WKHQ IROORZV WKDW (^Qf` S7 f DQG 9DUQff SY(^*Q7f` SY?7?+(^*Qf` SY?7?+ f )URP f f DQG f ZH FRQFOXGH WKDW 6Qf WKH LQFUHPHQW RI $Wf LV DOVR DQ I*Q ZLWK +XUVW SDUDPHWHU + PHDQ S7 DQG YDULDQFH SY?7?+ )URP 3URSRVLWLRQ

PAGE 73

LQ &KDSWHU Qf SRVVHVVHV /5' DV + 1RZ OHW & EH WKH IL[HG VHUYLFH UDWH LH WKH OLQN VSHHG RI DQ RXWSXW SRUW WKHQ WKH QXPEHU RI DUULYDOV XS WR WLPH W Wf LV HTXDO WR &W 'HQRWLQJ :Wf WKH QXPEHU RI FHOOV LQ WKH TXHXH XS WR WLPH W WKHQ E\ f5HLFKfV )RUPXODf >@ ZH KDYH :Wf VXS>DLf f§ cWf f§ D[f I[ff [W VXS>DLf f§ D[f f§ &W f§ [f@ [W VXS >S&fW[f \YS%+Wfa %+[fff f [W 7KH IROORZLQJ SURSRVLWLRQV DUH H[WHQGHG IURP WKH WKHRUHPV RULJLQDOO\ SUHn VHQWHG LQ >@ WR IDFLOLWDWH WKH GHULYDWLRQ RI WKH FHOO ORVV SUREDELOLW\ IRU D TXHXHLQJ V\VWHP ZLWK I*Q DUULYDO SURFHVV 3URSRVLWLRQ :Wf LV D VHOIVLPLODU SURFHVV RQO\ LI + f§ 3URRI :DWf f§ VXS>S f§ &f DW f§ D[f \YMO%+DWf f§ %QD[ff? [W VXS>D+S &fDOa+W [f AYMO%+Wf %+[ff? [W D+ VXS>DWf f§ D[f f§ >S f§ S f§ &fD`a+@W f§ [f@ f [W $V + f§ ZH REWDLQ :DWf DVXS>RWf f§ D[f f§ &W f§ [f@ [ W DO:Wf ZKLFK LPSOLHV WKDW : Wf LV D VHOIVLPLODU SURFHVV ZLWK + f§ ’ 1RWH WKDW IURP f ZKHQ + A :DWf WKH QXPEHU RI FHOOV LQ WKH TXHXH XS WR WLPH DW KDV WKH VDPH GLVWULEXWLRQ DV D+ WLPHV WKH QXPEHU RI FHOOV LQ D TXHXH ZLWK WKH RULJLQDO DUULYDO SURFHVV EXW ZLWK VHUYLFH UDWH S f§ S f§ &fD[a+

PAGE 74

3URSRVLWLRQ /HW S f§ T UHSUHVHQW WKH XWLOL]DWLRQ UDWH N WKH ILQLWH EXIIHU VL]H WKHQ A-I+ ‘ &nn:N + FRQVWDQW 3URRI 6LQFH $Wf KDV VWDWLRQDU\ LQFUHPHQWV :Wf LV VWDWLRQDU\ DQG WKXV KDV WKH VDPH GLVWULEXWLRQ DV :f 6XEVWLWXWLQJ W ZLWK LQ f ZH REWDLQ 93f VXS>L &ff§[f \YIL%+[ff f [ 6XEVWLWXWLQJ [ ZLWK W DQG ZLWK WKH IDFW WKDW %ZUf LV *DXVVLDQ ZH KDYH 3f VXS>S&fWf \YIL%+Wf@ W! R PD\>\IZ%+^Wf & SfW@ f /HW S^Nf EH WKH FHOO ORVV SUREDELOLW\ IRU D TXHXH ZLWK ILQLWH EXIIHU VL]H N WKHQ ZH KDYH S.f 3U:Wf !Nf 3UE8f Nf 3UPD[>A-,Y%+Wf f§ & f§ SfW@ Nf f 3UPD[>%fLf A f f 6LQFH %XWf LV D VHOIVLPLODU SURFHVV ZLWK +XUVW SDUDPHWHU + ZH KDYH Da+ %K DWf %Q Wf 6XEVWLWXWLQJ D ZLWK E+ ZH REWDLQ EO%+EOn+Wfs%+^Wf RU DIWHU FKDQJLQJ YDULDEOHV EO%+^Wf %+^EaO+Wf f /HW Mf GHQRWH 3UPD[>%+Wf f§ DW@ Ef 8VLQJ f ZH KDYH I 3UPD[>+Lf f§ DW@ Ef

PAGE 75

3UPD[>B"f f§ Ea[DW? f 3UPD[>%+^EaOA+Wf f§ Ea[DW@ f 3UPD[>%+Wf f§ EaA+DW@ f f &RPELQLQJ f DQG f ZH REWDLQ SNf 3UPD[>%fLf &ff9meBLf@ Mf f 3UPD[>%Wf f§ XW@ f f 0 f ZKHUH X! ‘ 6LQFH %QLIf LV *DQVVLDQ IURP f LW LV REYLRXV WKDW IX!f LV D PRQRWRQLFDOO\ GHFUHDVLQJ IXQFWLRQ 7KHUHIRUH JLYHQ D FHOO ORVV SUREDELOLW\ VD\ H WKHUH H[LVWV DQ LQYHUVH IXQFWLRQ VXFK WKDW X! IaO Hf FRQVWDQW 'HQRWLQJ S ZH KDYH 8,+ $S \Ir1 + ’ 3URSRVLWLRQ *LYHQ D IUDFWLRQDO TXHXHLQJ V\VWHP ZLWK 0&5 S YDULDQFH FRHIILFLHQW Y +XUVW SDUDPHWHU + VHUYLFH UDWH & DQG ILQLWH EXIIHU VL]H N WKH DSSUR[LPDWLRQ RI WKH FHOO ORVV SUREDELOLW\ SNf LV SNf H[S &Sf + f§ + SY++L +\Q\ ff f 3URRI

PAGE 76

)URP f ZH KDYH S^Nf 3U:ef Nf 3UPD[>%+Wf W@ f§S ]f f! 9r1 9r1 3UPD[>ef f§ f§f§f§ @ f Y W!R / Z f 6LQFH %+Wf W+%K f VXEVWLWXWLQJ %+Wf ZH KDYH 3^Nf 3UPD[>+Of & f§ \fW N W+9r1 !f 3Ue+Of QKQ> & QfW N W+9cA 4PLQ> A Y Q / & f§ QfW N W+9r1 @f ZKHUH WKH 4IXQFWLRQ LV GHILQHG E\ 4^;f a 9/ GW )RU EUHYLW\ GHQRWH E\ IWf :H KDYH +f^& QfW+ 9Lr1 B W+a??+f^&SfWN+f 9Lr1 f f f QR +^ +f& +^+ OfIFe L+ 9r1 f :KHQ W N+ er nerf DQG QR N+f+a> )f& $fSIIIFIIf +^+ fNf 9r1 N+f+a> +f& DLf@++Nf 9r1 f 7KHUHIRUH IWf KDV D PLQLPXP ZKHQ W Wr 6XEVWLWXWLQJ W ZLWK Wr DQG XVLQJ WKH DSSUR[LPDWLRQ 4^[f Ha[ DQG WKH WKHRU\ RI ODUJH GHYLDWLRQV >@ ZH REWDLQ & 9LWF9IL Kf
PAGE 77

f f *LYHQ SNf e WKH EDQGZLGWK & UHTXLUHG WR DFFHSW D FRQQHFWLRQ FDQ EH FRPSXWHG IURP f DQG LV & S +f§SY?QHfA+? A 9 O+? f§ + f 3URSRVHG &DOO $GPLVVLRQ &RQWURO $OJRULWKP 7KH VLPSOHVW &RQQHFWLRQ $GPLVVLRQ &RQWURO VWUDWHJ\ FDQ EH REWDLQHG FRQn VLGHULQJ 3HDN &HOO 5DWH 3&5f DOORFDWLRQ IRU HDFK WUDIILF VRXUFH 3&5 FDQ OHDG WR QR ORVV RI FHOOV EXW WR D YHU\ SRRU OLQN XWLOL]DWLRQ 7R REWDLQ D EHWWHU XWLOL]DWLRQ RI UHVRXUFHV LW LV QHFHVVDU\ WR DFFHSW D VPDOO GHJUDGDWLRQ RI WKH 4XDOLW\ RI 6HUYLFH 4R6f ZKLOH WDNLQJ LQWR DFFRXQW WKH VWDWLVWLFDO SURSHUWLHV RI DFWXDO WUDIILFV 7R WKLV HQG ZH SURSRVH D &DOO $GPLVVLRQ &RQWURO &$&f DOJRULWKP EDVHG RQ f 7KH SURFHGXUH RI WKH SURSRVHG &$& DOJRULWKP LV GHVFULEHG EHORZ (DFK VRXUFH KDV WR VXEPLW WKH WUDIILF SDUDPHWHUV LQFOXGLQJ L Y DQG + DQG GHVLUHG &HOO /RVV 5DWLR H &RPSXWH UHTXLUHG EDQGZLGWK & DFFRUGLQJ WR f &RPSDUH UHTXLUHG EDQGZLGWK & ZLWK DYDLODEOH EDQGZLGWK & $FFHSW WKH FRQQHFWLRQ LI & & RU UHMHFW RWKHUZLVH 1XPHULFDO 5HVXOWV DQG 'LVFXVVLRQV $VVXPH WKDW WKH RIIHUHG WUDIILF VRXUFH KDV WKH SDUDPHWHU XYf f 7R VWXG\ WKH LQIOXHQFH RI EXIIHU VL]H RQ FHOO ORVV SUREDELOLW\ ZH VHW WKH DOORFDWHG EDQGZLGWK & WR EH FHOOVHF )LJXUH GHSLFWV WKH FHOO ORVV SUREDELOLW\ YV EXIIHU VL]H IRU VHYHUDO +XUVW SDUDPHWHUV + ,W LV VHHQ WKDW ZKHQ + RU + WKH

PAGE 78

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f af 3&5 FHOOVVHF DQG WKH RXWSXW SRUW LV D '6 OLQN DW WKH VSHHG RI FHOOVVHF 7KH QXPEHU RI DGPLWWHG FRQQHFWLRQV YV EXIIHU VL]H LV SORWWHG LQ )LJXUH ZKHUH WKH QXPEHU RI DGPLWWHG FRQQHFWLRQV GHFUHDVHV DV WKH +XUVW SDUDPHWHU LQFUHDVHV 7KLV LV EHFDXVH WKH JUHDWHU WKH +XUVW SDUDPHWHU WKH EXUVWLHU WKH WUDIILF LV DQG FRQVHTXHQWO\ WKH ODUJHU DOORFDWHG EDQGZLGWK LV UHTXLUHG

PAGE 79

+ Hf af + Hf f + Hf f JHQXLQH + HVWLPDWHG &/3 JHQXLQH + HVWLPDWHG &/3 JHQXLQH + HVWLPDWHG &/3 ra ra ra r r ra r r ra r r r r r ra r ra ra r r r r ra r r r ra r rf§ rf§ 7DEOH 5REXVWQHVV RI WKH SURSRVHG &$& VFKHPH $VVXPH D ILQLWH EXIIHU VL]H FHOOV DQG WKH SDUDPHWHU X Yf f )LJXUH VKRZV WKH QXPEHU RI DGPLWWHG FRQQHFWLRQV JLYHQ D FHUWDLQ FHOO ORVV SUREn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f ZKLFK LV DQ XSSHU ERXQG RI WKH DFFXUDWH YDOXH ,I WKH SHDN UDWH DOORFDWLRQ VFKHPH LV XVHG DV WKH &$& WKHQ WKH QXPEHU RI DGPLWWHG FRQQHFWLRQV LV HTXDO WR WKH UDWLR RI WKH OLQN FDSDFLW\ WR WKH 3&5 $FFRUGn LQJO\ WKH QXPEHU RI DGPLWWHG FRQQHFWLRQV LQ WKLV FDVH LV )URP )LJXUH LW LV FOHDUO\ WKDW WKH SURSRVHG &$& DOJRULWKP DFKLHYHV PRUH DGPLWWHG FRQQHFWLRQV /HW XV GHILQH WKH VWDWLVWLFDO PXOWLSOH[LQJ JDLQ 60*f DV WKH UDWLR RI WKH QXPEHU RI DGPLWWHG FRQQHFWLRQV E\ XVLQJ WKH SHDN UDWH DOORFDWLRQ WR WKDW E\ XVLQJ

PAGE 80

JHQXLQH +XUVW SDUDPHWHU )LJXUH (VWLPDWHG &/3 YV JHQXLQH +XUVW SDUDPHWHUV WKH SURSRVHG DOJRULWKP 7KH 60* UHDFKHV D PD[LPXP YDOXH DV + :KHQ + WKH 60* LQFUHDVH IURP DV H WR DV H a :H FRQGXFWHG VLPSOH VLPXODWLRQV WR YHULI\ WKH JXDUDQWHH RI 4R6 LH WKH FHOO ORVV SUREDELOLW\ &/3f 7KH I*Q ZLWK +XUVW SDUDPHWHU EXIIHU VL]H FHOOV DQG X Yf f LV XVHG DV WKH WUDIILF VRXUFH 7KH DOORFDWHG EDQGZLGWK LV REWDLQHG IURP f 7KH VLPXODWLRQ UHVXOWV DUH VXPPDUL]HG LQ 7DEOH ZKHUH WKH ILUVW FROXPQ LV WKH UHTXHVWHG &/3 UDQJLQJ IURP a WR a DQG WKH VHFRQG DQG WKH WKLUG FROXPQ VKRZ WKH HVWLPDWHG &/3 E\ VLPXODWLRQ IRU VHOHFWHG &$& VFKHPHV &OHDUO\ WKH SURSRVHG &$& 3&$&f HQVXUHV WKH 4R6 +RZHYHU ZLWK WKH RWKHU VWDWLVWLFDO DOORFDWLRQ DOJRULWKP (TXLYDOHQW &DSDFLW\ (&f WKH 4R6 LV YLRODWHG PDLQO\ EHFDXVH RI WKH XVH RI DQ LQDSSURSULDWH WUDIILF PRGHO ZKLFK IDLOV WR FDSWXUH WKH FKDUDFWHULVWLFV RI VHOIVLPLODULW\

PAGE 81

UHTXHVWHG &/3 HVWLPDWHG &/3 ZLWK 3& $& HVWLPDWHG &/3 ZLWK (& 7 r ra 7 raV r 7 ra r &7 rf§ r f ra ra LFU rf§ r f rr r 7DEOH *XDUDQWHH RI 4R6 D &' R R 7 &' ( ‘D n ] n %XIIHU VL]H n )LJXUH 1XPEHU RI DGPLWWHG FRQQHFWLRQV YV EXIIHU VL]H

PAGE 82

)LJXUH 1XPEHU RI DGPLWWHG FRQQHFWLRQV YV FHOO ORVV SUREDELOLW\ &RQFOXVLRQ ,Q WKLV &KDSWHU ZH SURSRVHG D SUHYHQWLYH IORZ FRQWURO VFKHPH FRQVLVWLQJ RI D &RQQHFWLRQ $GPLVVLRQ &RQWURO &$&f WDNLQJ LQWR DFFRXQW WKH +XUVW SDUDPHWHU RI LQFRPLQJ WUDIILF VRXUFHV 7KH &$& LV LPSOHPHQWHG EDVHG RQ DQ XSSHU ERXQG RI WKH FHOO ORVV SUREDELOLW\ IRU WKH VHOIVLPLODU TXHXHLQJ PRGHO ZLWK D I*Q DUULYDO SURFHVV 7KH QXPHULFDO UHVXOWV VKRZHG WKDW ZLWK WKH SURSRVHG &$& PRUH FRQQHFWLRQV FDQ EH DFFHSWHG LQWR WKH QHWZRUNV WKDQ ZLWK SHDN UDWH DOORFDWLRQ &$& ,Q WKH SURSRVHG &$& HDFK VRXUFH KDV WR VXEPLW WKH WUDIILF SDUDPHWHUV LQn FOXGLQJ S Y DQG + DQG GHVLUHG &HOO /RVV 5DWLR H 7KLV &$& LV WKHUHIRUH EHWWHU VXLWHG IRU DSSOLFDWLRQV VXFK DV 9LGHR RQ 'HPDQG 92'f 2Q WKH RWKHU KDQG IRU WKH &$& WR EH DEOH WR VXSSRUW UHDOWLPH WUDIILF WKHVH SDUDPHWHUV KDYH WR EH FRPSXWHG RQ WKH IO\ ,W SRVHV D FKDOOHQJH HVSHFLDOO\ IRU HVWLPDWLQJ WKH +XUVW SDUDPHWHU ZKLFK UHTXLUHV ORQJHU WLPH LQWHUYDO WR REWDLQ PRUH DFFXUDWH HVWLPDWLRQ

PAGE 83

&+$37(5 3,' )/2: &21752/ 6&+(0( )25 9%5 75$)),& ,Q FRPPXQLFDWLRQ QHWZRUNV ZLWK ODUJH GHOD\EDQGZLGWK SURGXFW FRQJHVWLRQ FRXOG KDSSHQ RYHU VKRUWHU WLPH VFDOHV WKDQ WKRVH DW ZKLFK HQGWRHQG SURWRFROV VXFK DV FRQJHVWLRQ FRQWURO VFKHPHV W\SLFDOO\ RSHUDWH ,Q VXFK FDVHV WKH FRQJHVWLRQ FDQ GLVVLSDWH UDSLGO\ EHIRUH FRQJHVWLRQ IHHGEDFN LQIRUPDWLRQ UHWXUQV WR WKH VRXUFH 1HWn ZRUN GHVLJQHUV WKHUHIRUH IDFH D FKDOOHQJH 7KH EXUVW\ DQG F\FOLF QDWXUH RI 9DULDEOH %LW 5DWH 9%5f WUDIILF FUHDWHV DQRWKHU LVVXH IRU WUDQVPLVVLRQ LQ $70 QHWZRUNV 7R UHDFK WKH GXDO JRDOV RI NHHSLQJ FHOO ORVV UDWH ORZ DQG QHWZRUN XWLOL]DWLRQ KLJK ZH SURn SRVH DQ DGDSWLYH UDWHEDVHG IORZ FRQWURO VFKHPH IRU UHDOWLPH 9%5 WUDIILF LQ $70 QHWZRUNV 7KH JRDO RI WKH VFKHPH LV WR PLQLPL]H WKH LPSDFW RI WUDIILF RYHUORDG LQ RUGHU WR OLPLW WKH FHOO ORVV UDWH WR DQ DFFHSWDEOH UDQJH DQG DOVR LQFUHDVH WKH QHWZRUN XWLn OL]DWLRQ 7KH SURSRVHG IORZ FRQWURO VFKHPH LV EDVHG RQ SUHGLFWLQJ WKH HYROXWLRQ RI EXIIHU RFFXSDQF\ RYHU WLPH XVLQJ D 3URSRUWLRQDOSOXV,QWHJUDOSOXV'HULYDWLYH 3,'f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

PAGE 84

,QWURGXFWLRQ $V\QFKURQRXV 7UDQVIHU 0RGH $70f LV ZLGHO\ FRQVLGHUHG WR EH WKH QH[W JHQn HUDWLRQ RI KLJK VSHHG LQWHUQHWZRUNLQJ WHFKQRORJ\ PDLQO\ GXH WR WKH VFDODEOH EXLOGLQJ EORFNV LW SURYLGHV WR VHUYH DV D VWDEOH IRXQGDWLRQ IRU HYROYLQJ SXEOLF DQG SULYDWH QHWZRUNV :LWK WKH DGYHQW RI %URDGEDQG ,6'1 $70 QHWZRUNV PXVW VXSSRUW DSSOLn FDWLRQV ZLWK GLYHUVH WUDIILF FKDUDFWHULVWLFV DQG TXDOLW\ RI VHUYLFH 4R6f UHTXLUHPHQWV 7KHUHIRUH DQ DFFXUDWH WUDIILF FKDUDFWHUL]DWLRQ LV RI SULPDU\ LPSRUWDQFH LQ LPSOHPHQWn LQJ HIIHFWLYH IORZ FRQWURO VWUDWHJLHV DQG PDNLQJ HIILFLHQW XVH RI QHWZRUN UHVRXUFHV LQ $70 QHWZRUNV 1HWZRUN WUDIILF LV XVXDOO\ FKDUDFWHUL]HG E\ WKH WROHUDQFH RI LQIRUPDWLRQ GHOD\ RU ORVV )RU H[DPSOH WKH WUDIILF IRU $YDLODEOH %LW 5DWH $%5f RU EHVWHIIRUW VHUYLFH GHPDQGV VXSHULRU &HOO /RVV 5DWH &/5f EXW \LHOGV LQIHULRU SHUIRUPDQFH LQ WHUPV RI &HOO 7UDQVIHU 'HOD\ &7'f DQG &HOO 'HOD\ 9DULDWLRQ &'9f 2Q WKH RWKHU KDQG UHDOn WLPH WUDIILF VXFK DV WKDW IRU &RQVWDQW %LW 5DWH &%5f VHUYLFH LV GLVWLQJXLVKHG E\ LWV VWULQJHQW &7' DQG &'9 UHTXLUHPHQWV 5HDOWLPH WUDIILF FRQWUDVWV VLJQLILFDQWO\ IURP HODVWLF RU EHVWHIIRUW WUDIILF E\ SULPDULO\ VXSSRUWLQJ DSSOLFDWLRQV WKDW KDYH GHGLFDWHG EDQGZLGWK 7KHVH DSSOLFDWLRQV KRZHYHU GR KDYH VRPH DELOLW\ WR DGDSW WR QHWZRUN FRQGLWLRQV VXFK DV FHOO GHOD\ RU FHOO ORVVHV EXW RQO\ ZLWKLQ D PDUJLQ RI WROHUDQFH >@ 7KH RWKHU W\SH RI UHDOWLPH WUDIILF LQ DGGLWLRQ WR &%5 WUDIILF LV 9DULDEOH %LW 5DWH 9%5f WUDIILF VXFK DV WKDW JHQHUDWHG LQ YLGHRFRQIHUHQFLQJ XVLQJ FRPSUHVVLRQ 7KH FKDUDFWHULVWLFV RI 9%5 YLGHR WUDIILF GHSHQGV SULPDULO\ RQ WZR IDFWRUV WKH QDWXUH RI WKH YLGHR QDPHO\ ZKHWKHU WKHUH DUH IUHTXHQW RU UDUH VFHQH FKDQJHV WKH FRGLQJ VFKHPH HPSOR\HG VXFK DV IRU H[DPSOH 03(* + HWF )RU WHOHFRQIHUHQFLQJ DQG YLGHRWHOHSKRQH DSSOLFDWLRQV WKH &&,77 + VWDQn GDUG VSHFLILHV FRPSUHVVLRQ WHFKQLTXHV DW UDWHV RI S[ NLORELWVVHFRQG ZKHUH S UDQJHV IURP WR DERXW $QRWKHU ZHOONQRZQ FRPSUHVVLRQ DOJRULWKPV IRU YLGHR LV 03(*

PAGE 85

0RWLRQ 3LFWXUHV ([SHUW *URXS GHILQHG E\ ,62 &RPSDUHG WR + WKH 03(* VWDQGDUG ZDV GHVLJQHG IRU D KLJKHU UDQJH RI UDWHV DQG PXFK EHWWHU YLVXDO TXDOLW\ $OWKRXJK 03(* YLGHR LV VXLWDEOH IRU D ODUJH QXPEHU RI PXOWLPHGLD DSSOLFDWLRQV LQFOXGLQJ YLGHRFRQIHUHQFLQJ GLVWDQW OHDUQLQJ DQG YLGHR PDLO LW LV QRW KRZHYHU LQWHQGHG WR EH EURDGFDVW WHOHYLVLRQ TXDOLW\ ZKLFK VLJQDOV DW 0ESV $Q 03(* FRGHU JHQHUDWHV WKUHH W\SHV RI IUDPHV ,QWUDIUDPHV ,f WKDW XVH LPDJH FRPSUHVVLRQ VFKHPH 3UHGLFWHG 3f DQG %LGLUHFWLRQDO %f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f VKRZV WKH IUDPH VL]H RYHU WLPH HODSVHG RI D W\SLFDO 03(* FRPn SUHVVHG YLGHR ZLWK I? HTXDO WR *23 f REWDLQHG IURP D UHFRUGHG VHTXHQFH RI 79 DGYHUWLVHPHQWV )LJXUH Ef GHSLFWV WKH UHODWLYH IUHTXHQF\ RI WKH IUDPH VL]H 7KH HQFRGLQJ SDWWHUQ LV DV IROORZV ,%%%%%%%%%3%%%%%%%%%3%%%%%%%%%,%%

PAGE 86

Df IUDPH VL]H FHOOV )LJXUH 03(* IUDPH VL]H DQG LWV SUREDELOLW\ GHQVLW\ IXQFWLRQ ,W LV QRWHG WKDW EHWZHHQ WZR VXFFHVVLYH IUDPHV D 3 IUDPH LV IROORZHG E\ FRQVHFXWLYH % IUDPH $V )LJXUH Ef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

PAGE 87

*HQHULF 9LGHR 7UDQVPLVVLRQ 6\VWHP ZLWK )HHGEDFN $PRQJ DOO WUDIILF FODVVHV LQ $70 QHWZRUNV UHDOWLPH YLGHR WUDIILF SRVHV D XQLTXH FKDOOHQJH 6LQFH WKH UHTXLUHG VHUYLFH LV GHOD\ VHQVLWLYH WKH QHWZRUN PXVW SURn YLGH D UHVRXUFH UHVHUYDWLRQ VFKHPH WR DOORFDWH QHWZRUN UHVRXUFHV IRU HDFK 9%5 YLGHR VWUHDP +RZHYHU WKH EXUVWLQHVV RI 9%5 YLGHR WUDIILF PDNHV LW HVSHFLDOO\ GLIILFXOW WR GHWHUPLQH WKH DPRXQW RI UHVRXUFHV UHTXLUHG 2Q WKH RQH KDQG LI UHVRXUFHV DUH UHVHUYHG DFFRUGLQJ WR WKH 6XVWDLQDEOH &HOO 5DWH 6&5f RI WKH 9%5 YLGHR VRXUFHV XQn DFFHSWDEOH GHOD\V DQG FHOO ORVVHV PD\ UHVXOW ZKHQ WKH YLGHR LV WUDQVPLWWHG DW WKH SHDN UDWH 2Q WKH RWKHU KDQG LI UHVRXUFHV UHVHUYDWLRQV DUH EDVHG RQ WKH 3HDN &HOO 5DWH 3&5f RI WKH YLGHR VRXUFHV WKH QHWZRUN FRXOG EH XQGHUXWLOL]HG PRVW RI WKH WLPH 2QH VWUDWHJ\ WR EHVW FRPSURPLVH ERWK SUREOHPV LV WR G\QDPLFDOO\ DGMXVW WKH VHQGLQJ UDWH RI ELWVWUHDPV DFFRUGLQJ WR WKH QHWZRUN LQIRUPDWLRQ REWDLQHG YLD D IHHGEDFN SDWK EHWZHHQ WKH QHWZRUN DQG WKH WUDQVPLWWHU )LJXUH GHSLFWV D JHQHULF YLGHR WUDQVPLVVLRQ V\VWHP ZKHUH D YLGHR VLJQDO LV IHG WR D YLGHR HQFRGHU 7KH HQFRGHG ELWVWUHDP LV SURGXFHG E\ YLGHR FRPSUHVVLRQ FRGHFV VXFK DV 03(* RU + ZLWK DQ DWWHPSW WR FRPSUHVV WKH YLGHR VWUHDP E\ UHPRYLQJ VSDWLDO DQG WHPSRUDO UHGXQGDQFLHV 7KH UDWH FRQWUROOHU LV DEOH WR VHOHFW WKH WUDQVPLVVLRQ UDWH E\ DGMXVWLQJ WKH TXDQWL]DWLRQ OHYHO DFFRUGLQJ WR IHHGEDFN LQIRUPDn WLRQ 7KLV IHHGEDFN SDWK KRZHYHU LQ VRPH VWDQGDUG VXFK DV + KDV DQ RSWLRQ WR GLVDEOH LW 7KLV HQDEOHV XVHUV WR PD[LPL]H TXDOLW\ DW WKH H[SHQVH RI EDQGZLGWK 6HYHUDO IORZ FRQWURO VWUDWHJLHV IRU PDQDJLQJ RYHUORDG RI 9%5 YLGHR WUDIILF DUH GHVFULEHG DV IROORZV f ,QWHJUDWHG UDWHEDVHG IORZ FRQWURO :HL DQG HW DO >@ SURSRVHG D MRLQW HQn FRGHU DQG FKDQQHO UDWH FRQWURO DOJRULWKP IRU $70 QHWZRUNV ZLWK OHDN\ EXFNHWV DV RSHQORRS VRXUFH IORZ FRQWURO PRGHOV WR EDODQFH ERWK LVVXHV RI FRQVLVWHQW YLGHR TXDOLW\ RQ WKH HQFRGHU VLGH DQG ELWVWUHDP VPRRWKQHVV IRU WKH VWDWLVWLn FDO PXOWLSOH[LQJ JDLQ 60*f RQ WKH QHWZRUN VLGH 7KH DOJRULWKP FRQVLGHUV

PAGE 88

9LGHR VHQGHU 9LGHR UHFHLYHU )LJXUH *HQHULF 9LGHR 7UDQVPLVVLRQ 6\VWHP ZLWK )HHGEDFN FRQVWUDLQWV LPSRVHG E\ WKH HQFRGHU DQG GHFRGHU EXIIHUV WKH OHDN\ EXFNHW FRQn WURO WUDIILF VPRRWKLQJ DQG UDWH FRQWURO 7KH HQFRGHU UDWH FRQWURO LV VHSDUDWHG LQWR D VXVWDLQDEOHUDWH FRQWURO DQG D XQLGLUHFWLRQDO LQVWDQWDQHRXVUDWH FRQWURO /HH DQG HW DO >@ VXJJHVWHG D PRGLILHG H[SOLFLW UDWH LQGLFDWLRQ IRU FRQJHVn WLRQ DYRLGDQFH 0(5,&$f VFKHPH IRU ERWK 9%5 DQG &%5 WUDIILFV LQ $70 QHWZRUNV 7KLV PRGLILHG VFKHPH EHORQJV WR WKH FODVV RI FORVHGORRS UDWHEDVHG FRQJHVWLRQ FRQWURO VFKHPH f $GDSWLYH FUHGLWEDVHG IORZ FRQWURO *R\DO DQG HW DO >@ SUHVHQWHG D UHFHLYHU RULHQWHG DGDSWLYH FUHGLWEDVHG IORZ FRQWURO VFKHPH WR PLQLPL]H WKH EXIIHU UHn TXLUHPHQW LQ WKH QHWZRUN ZKLOH JXDUDQWHHLQJ WKDW FHOOV RI 9%5 HQFRGHG YLGHR IORZV ZLOO QRW EH ORVW :LWK WKH SURSRVHG EDQGZLGWK HVWLPDWLRQ WHFKQLTXHV ZKLFK H[SORLW WKH VWUXFWXUH RI WKH YLGHR WUDIILF LW DOVR PLQLPL]HV WKH HQGWRHQG GHOD\ DQG MLWWHU IRU 9%5 HQFRGHG YLGHR VWUHDPV f 'DWD UHFRYHU\ /RVW GDWD PD\ EH UHFRYHUHG XVLQJ WHPSRUDO DQG VSDWLDO LQWHUSRn ODWLRQ RU IRUZDUG HUURU UHFRYHU\ PHWKRGV +X DQG /LQ >@ VXJJHVWHG D IRUZDUG HUURU FRQWURO VFKHPH IRU 03(* YLGHR WR UHFRYHU ORVW FHOOV LQ $70 QHWZRUNV 7KH RULJLQDO 03(* VWUHDPV DUH VSOLW LQWR Q VXEVWUHDPV DQG WKH VXEVWUHDPV DUH

PAGE 89

VNHZHG E\ D IL[HG DPRXQW RI WLPH IURP HDFK RWKHU 7KH VXEVWUHDPV DUH WUDQVn PLWWHG ZLWK WKH IRUZDUG HUURU FRUUHFWLRQ FRGH SURYLGLQJ D GHWHUPLQLVWLF ERXQG RQ WKH GHOD\ DQG KDYLQJ WKH LPSOHPHQWDWLRQ VLPSOLFLW\ QHHGHG IRU KLJKVSHHG YLGHR DSSOLFDWLRQV f 3ULRULWL]HG WUDQVPLVVLRQ ,VPDLO DQG HW DO >@ SURSRVHG D SULRULW\ VFKHPH IRU 03(* YLGHR ZKLFK LV LPSOHPHQWHG LQ D VRIWZDUH HQFRGHU WKDW SURGXFHV D SURn SRUWLRQDWH WUDIILF LQ ERWK LH KLJK DQG ORZf SULRULW\ SDUWLWLRQV IRU DOO WKUHH IUDPH W\SHV XVHG LQ 03(* $Q $70 PXOWLSOH[HU ZLWK D SXVKRXW EXIIHU VFKHPH LV LPSOHPHQWHG WR SURYLGH SULRULW\ VFKHGXOLQJ DW WKH PXOWLSOH[HU IRU WKH WZR SULn RULW\ SDUWLWLRQV f 7UDIILF VKDSLQJ DQG VPRRWKLQJ %XUVWLQHVV RI HQFRGHG WUDIILF PD\ EH UHPRYHG E\ VPRRWKLQJ LW RYHU DQ LQWHUYDO +HUH RQH KDV WR PDNH D WUDGHRII EHWZHHQ FHOO ORVVHV GXH WR EXUVWLQHVV DQG DOORZHG GHOD\ /L DQG HW DO >@ VXJJHVWHG D QHZ IORZ FRQWURO IXQFWLRQ FDOOHG fWLPHGULYHQf SULRULW\ ZKLFK LV DQ LQWHUQDO WUDIILF VKDSLQJ PHFKDQLVP VXSSRUWLQJ &%5 ZLWK GHWHUPLQLVWLF JXDUDQWHHV DQG 9%5 ZLWK VWDWLVWLFDO PXOWLSOH[LQJ 7KH PHFKDQLVP GRHV QRW UHTXLUH WKH LGHQWLILFDWLRQ DQG VHSDUDWLRQ RI SDFNHW IORZV RI GLIIHUHQW UHDOWLPH VHVVLRQVFRQQHFWLRQV LQVLGH WKH QHWZRUN f $GPLVVLRQ FRQWURO 7KLV PHFKDQLVP FDQ EH XVHG WR DFFHSW RU UHMHFW QHZ FRQn QHFWLRQV DW WKH FRQQHFWLRQ VHWXS SKDVH $70 QHWZRUNV UHTXLUH WUDIILF FKDUn DFWHULVWLFV LQIRUPDWLRQ VXFK DV &/5 3&5 6&5 DQG &'9 IRU 9%5 WUDIILF WR JXDUDQWHH WKH 4R6 RI D FRQQHFWLRQ 3URSRVHG )ORZ &RQWURO 6FKHPH 7KH SURSRVHG IORZ FRQWURO VFKHPH LV FRPSDUDEOH WR D FODVVLFDO WUDFNLQJ FRQWURO V\VWHP LQ ZKLFK DSSURSULDWH IHHGEDFN LV DSSOLHG WR KDYH WKH V\VWHP RXWSXW \>Q@ WUDFN WKH LQSXW U>Q@ DV )LJXUH Df VKRZV $ FRPSHQVDWRU WDNLQJ WKH GLIIHUHQFH EHWZHHQ

PAGE 90

Df U>Q@ t &RPSHQVDWRU 3ODQW \>Q@ Ef Er  &RQWUROOHU 1HWZRUN )LJXUH *HQHULF WUDFNLQJ V\VWHP DQG LWV DSSOLFDWLRQ LQ QHWZRUNV WKH UHIHUHQFH U>Q@ DQG RXWSXW \>Q@ DV D WULJJHU JHQHUDWHV D G\QDPLFDO VLJQDO HQWHULQJ WKH SODQW 7KH FRPSHQVDWRU PXVW EH ZHOO GHVLJQHG VR \>Q@ FDQ WUDFN U>Q@ SURPSWO\ 2QH RI WKH PRVW ZHOONQRZQ FRPSHQVDWRUV LV SUREDEO\ WKH VRFDOOHG 3URSRUWLRQDO SOXV,QWHJUDOSOXV'HULYDWLYH 3,'f FRQWUROOHU )LJXUH Ef VKRZV WKH VFHQDULR WKDW DSSOLHV WKH WUDFNLQJ V\VWHP WR QHWZRUNV ,Q WKLV FKDSWHU ZH H[SORLW WKLV V\VWHP WKHRUHWLF PHWKRGRORJ\ LQ RXU SURSRVHG IORZ FRQWURO VFKHPH )LJXUH 3,' IORZ FRQWUROOHU

PAGE 91

,Q WKH SURSRVHG IORZ FRQWURO VFKHPH ZKRVH EORFN GLDJUDP VKRZQ LQ )LJXUH 5Q WKH PD[LPXP VHQGLQJ UDWH RI WKH QWK IUDPH LV GHWHUPLQHG E\ WKH HTXDWLRQ 5Q 5LQ >.S f EQ Erf .M  ELOEA .RLEQEQA5LQ f L Qf§O ZKHUH 5LQ LV WKH VHQGLQJ UDWH IRU WKH RULJLQDO ELWVWUHDP RI WKH QWK IUDPH ZLWKRXW IHHGEDFN EQ WKH EXIIHU RFFXSDQF\ DW WKH SHUIRUPDQFH ERWWOHQHFN DW WKH EHJLQQLQJ RI WKH WUDQVPLVVLRQ RI WKH QWK IUDPH EQ WKH HVWLPDWHG YDOXH RI EQ Er WKH GHVLUHG EXIIHU RFFXSDQF\ DQG O D QRQQHJDWLYH LQWHJHU ,Q (TXDWLRQ WKH DFWXDO VHQGLQJ UDWH 5Q LV XSGDWHG RQFH SHU IUDPH LQWHUYDO EDVHG RQ WKH EXIIHU RFFXSDQF\ REWDLQHG IURP WKH IHHGEDFN 7KH JRDO RI WKH FRQWURO DFWLRQ LV WR NHHS WKH EXIIHU RFFXSDQF\ OHYHO FORVH WR WKH GHVLUHG YDOXH Er 7KH VHQGLQJ UDWH 5Q LV DGMXVWHG E\ WKH FRQWUROOHU VR WKH EXIIHU RFFXSDQF\ LV GULYHQ WR Er SURPSWO\ 7KH HVWLPDWHG EXIIHU RFFXSDQF\ E LV REWDLQHG E\ XVLQJ D OLQHDU SUHGLFWRU GHVFULEHG E\ WKH IROORZLQJ HTXDWLRQ IFL EQ .N (QL f§ N f f S@ f WI f L ZKHUH J LV D FRQVWDQW VHUYLFH UDWH DW WKH EXIIHU RI LQWHUHVW EQN WKH ODWHVW UHSRUWHG EXIIHU RFFXSDQF\ IURP IHHGEDFN DW WKH EHJLQQLQJ RI WKH WUDQVPLVVLRQ RI WKH Q£IUDPH DQG WI LV WKH IUDPH LQWHUYDO LH WKH LQYHUVH YDOXH RI WKH IUDPH UDWH /HW HQ GHQRWH WKH YDOXH RI WKH HUURU VLJQDO FRUUHVSRQGLQJ WR WKH QWK IUDPH WKDW LV HQ EQ f§ Er $HQ WKH GLIIHUHQFH RI WZR FRQVHFXWLYH HUURU VLJQDOV DQG (}HQ=f WKH FXPXODWLYH HUURU VLJQDOV LQ O IUDPHV :H FDQ WKHQ UHZULWH (TXDWLRQ DV IROORZV 5Q 5LQ .S ‘ HQ .c ‘ HQ Of .' ‘ $HQf f 5LQ f 6LQFH IrHQ f HQ ZLWKRXW WKH ORVV RI JHQHUDOLW\ ZKHQ O (TXDWLRQ FDQ EH IXUWKHU VLPSOLILHG WR

PAGE 92

5Q 5LQ ^.L ‘ Hf Of .' f $HQf f 5LQ f &OHDUO\ ZH FDQ VHH WKDW IURP f WKH LQVWDQWDQHRXV VHQGLQJ UDWH RI WKH QWK IUDPH LV GHWHUPLQHG E\ .G D GHULYDWLYH IDFWRU .c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f FHOOV FDUU\LQJ EXIIHU LQIRUPDWLRQ DW WKH HJUHVV DGDSWHU DUH JHQHUDWHG SHULRGLFDOO\ DQG DUH VHQW EDFN WR DOO YLGHR VRXUFHV E\ FRQQHFWLRQV LQ WKH

PAGE 93

)LJXUH 1HWZRUN 6FHQDULR UHYHUVH GLUHFWLRQ 'XH WR WKH EDODQFHG QHWZRUN WRSRORJ\ VLPLODU QHWZRUN EHKDYLRU DW HDFK &3( LV H[SHFWHG IURP VLPXODWLRQV 1XPHULFDO 5HVXOWV DQG 'LVFXVVLRQV 7R HYDOXDWH WKH SHUIRUPDQFH RI WKH SURSRVHG 3,' IORZ FRQWURO VFKHPH XQGHU D FRQJHVWHG QHWZRUN FRQGLWLRQ HDFK 9& IRU YRLFH JHQHUDWHV WUDIILF FRQWULEXWLQJ b RI WKH EDQGZLGWK RI WKH WUXQN OLQN ZKLOH HDFK 9& IRU YLGHR LV FRQWHQGLQJ IRU WKH 7R &36 $ )URP &36 $ 9%5 f 9& IRU &%5 f 9& IRU 9%5 f§f§‘‘‘‘ 9& IRU )HHGEDFN )LJXUH )XQFWLRQDO PRGHO RI &3(

PAGE 94

UHPDLQLQJ b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f DQG )LJXUH Ef VKRZ WKH IUDPH VL]H WUDQVPLWWHG DW &3(7 DQG WKDW UHFHLYHG DW &3(' UHVSHFWLYHO\ 7KH &/5 &HOO /RVV 5DWHf GHILQHG E\ WKH UDWLR RI 17 WKH WRWDO QXPEHU RI FHOOV WUDQVPLWWHG WR 1/O WKH WRWDO QXPEHU RI FHOO ORVW FDQ EH REWDLQ IURP )LJXUH Df DQG )LJXUH Ef 7KH &/5 LQ WKLV FDVH LV $I\ 1/ f§ f 7KH &/5 LV FOHDUO\ OHVV WKDQ b ZLWKRXW XVLQJ WKH UDWH FRQWUROOHU EXW SA WKH QRUPDOL]HG QHWZRUN XWLOL]DWLRQ UDWH LV DV ORZ DV b ZKLFK PHDQV WKH EDQGZLGWK DYDLODEOH WR 9%5 WUDIILF LV XQXVHG b RI WKH WLPH )LJXUH DQG )LJXUH VKRZ WKH UHVXOWV IRU WKH VDPH VWDWLVWLFV DV WKRVH RI )LJXUH DQG )LJXUH ZKHQ WKH SURSRVHG IORZ FRQWURO VFKHPH LV XVHG ZLWK .Gf f 1RWLFH WKDW ZLWK PRUH FHOOV WUDQVPLWWHG DQG EHWWHU YLGHR TXDOLW\f WKDQ WKH FDVH RI QR IHHGEDFN WKH &/5 LQFUHDVHV WR $I\ 1/ f EXW S1 LV LPSURYHG VXEVWDQWLDOO\ WR b 7KH UHVXOWV XQGHU GLIIHUHQW YDOXHV RI .M .Tf DUH OLVWHG LQ 7DEOH %RWK &/5 DQG SA LQFUHDVH DV WKH YDOXHV RI .M.G LQFUHDVH $QRWKHU RSSRUWXQLW\ WR WDNH DGYDQWDJH RI WKH SURSRVHG IORZ FRQWURO VFKHPH DQG PDNH WKH PRVW XVH RI DYDLODEOH EDQGZLGWK LV WR DJJUHJDWH PRUH 9%5 VWUHDPV /HW 1YEU GHQRWH WKH QXPEHU RI 9&V IRU YLGHR DW HDFK &3( 7KH &/5 DQG S1 FRUUHVSRQGLQJ WR GLIIHUHQW YDOXH RI 1AU DUH VKRZQ LQ 7DEOH $V 1fEU LQFUHDVHV PRUH EXUVW\ YLGHR

PAGE 95

Df Ef )LJXUH 7UDQVPLWWHG DQG UHFHLYHG IUDPH VL]H ZLWK QR IHHGEDFN FRQWURO .M.Gf &/5 3Q f f f f f f f 7DEOH &/5 DQG QRUPDOL]HG XWLOL]DWLRQ UDWH XQGHU YDULRXV .f

PAGE 96

IO R WLPH VHFf [ Df Ef )LJXUH &XPXODWLYH QXPEHU RI FHOOV VHQW DQG ORVW ZLWK QR IHHGEDFN FRQWURO

PAGE 97

Df Ef )LJXUH 7UDQVPLWWHG DQG UHFHLYHG IUDPH VL]H ZLWK IHHGEDFN FRQWURO

PAGE 98

Ef )LJXUH &XPXODWLYH QXPEHU RI FHOOV VHQW DQG ORVW ZLWK IHHGEDFN FRQWURO

PAGE 99

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

PAGE 100

&+$37(5 ,17(*5$7(' 6(59,&(6 35272&2/6 $1' 75$)),& $1$/<6,6 7KH ,QWHUQHW KDV SURYHQ WR EH WKH EHVW VROXWLRQ IRU JOREDO VKDULQJ RI LQIRUn PDWLRQ 7KURXJK WKH ,QWHUQHW RQH FDQ DFFHVV LQIRUPDWLRQ IURP ZRUOG ZLGH ZHE :::f DQG H[FKDQJH HOHFWURQLF PDLO DV ZHOO DV VKDUH DSSOLFDWLRQV ZLWK RWKHUV IURP DQ\ FRQWLQHQW RQ HDUWK 7KH VXFFHVV RI WKHVH HQGWRHQG FRPPXQLFDWLRQV LV HVVHQn WLDOO\ DFKLHYHG E\ WZR WUDQVSRUW OD\HU SURWRFROV LQ WKH ,QWHUQHW SURWRFRO VXLWH QDPHO\ 7UDQVSRUW &RQWURO 3URWRFRO 7&3f DQG 8VHU 'DWDJUDP 3URWRFRO 8'3f ZKHUH 7&3 SURYLGHV D FRQQHFWLRQRULHQWHG UHOLDEOH IORZ EHWZHHQ WZR KRVWV DQG 8'3 SURYLGHV D FRQQHFWLRQOHVV XQUHOLDEOH GDWDJUDP VHUYLFH &XUUHQWO\ RQH FODVV RI VHUYLFH LQ WKH ,QWHUQHW H[LVWV QRUPDOO\ UHIHUUHG WR DV fEHVW HIIRUWf ZKRVH WUDIILF LV FKDUDFWHUL]HG E\ ILUVW FRPH ILUVW VHUYH VFKHGXOLQJ DW HDFK QRGH LQ WKH QHWZRUN %HVW HIIRUW VHUYLFH KDV ZRUNHG H[WUHPHO\ ZHOO IRU WUDIILF ZLWKRXW WLPH FRQVWUDLQW +RZHYHU IRU UHDOWLPH WUDIILF VXFK DV YRLFH DQG YLGHR 7&38'3 KDV SHUIRUPHG ZHOO RQO\ ZKHQ QHWZRUNV DUH XQGHUORDGHG ,Q RUGHU WR SURYLGH JXDUn DQWHHG TXDOLW\ RI VHUYLFH 4R6f IRU UHDOWLPH WUDIILF FHDVHOHVVO\ QHZ FODVVHV RI VHUYLFH DQG QHZ SURWRFROV DUH EHLQJ LQWURGXFHG RQ WKH ,QWHUQHW ZKLOH UHWDLQLQJ WKH H[LVWLQJ EHVW HIIRUW VHUYLFH 7KUHH HPHUJLQJ WHFKQRORJLHV WKDW ZLOO EH DEOH WR VXSSRUW WKH QHHGHG ,QWHUQHW PXOWLPHGLD VHUYLFHV DUH f5HVRXUFH 5HVHUYDWLRQ 3URWRFRO 5693ff f5HDO7LPH 7UDQVSRUW 3URWRFRO 573ff DQG f,3 0XOWLFDVW 3URWRFROf 1HWZRUN WUDFHV KDYH EHHQ ZLGHO\ FRQVLGHUHG DV RQH RI WKH PRVW HIILFLHQW ZD\V WR VWXG\ QHWZRUN G\QDPLFV XVDJH FKDUDFWHULVWLFV DQG JURZWK SDWWHUQV 7KH\ DUH RI H[WUHPH LPSRUWDQFH WR WUDFHGULYHQ VLPXODWLRQV IRU WRROV OLNH 231(7 ,Q WKLV &KDSn WHU ZH GHVFULEH WKH WKUHH UHDOWLPH RULHQWHG SURWRFROV PHQWLRQHG DERYH LQ 6HVVLRQ

PAGE 101

7UDIILF PHDVXUHPHQWV DQG DQDO\VLV EDVHG RQ WKH WUDFHV FROOHFWHG LQ WKH /DERUDWRU\ IRU ,QIRUPDWLRQ 6\VWHPV DQG 7HOHFRPPXQLFDWLRQV /,67f LV SUHVHQWHG LQ 6HVVLRQ 3URWRFROV IRU ,QWHJUDWHG 6HUYLFHV 5HVRXUFH 5HVHUYDWLRQ 3URWRFRO 5693 LV D UHVRXUFH UHVHUYDWLRQ VHWXS SURWRFRO GHVLJQHG IRU DQ LQWHJUDWHG VHUn YLFHV ,QWHUQHW 7KH 5693 PHFKDQLVPV SURYLGH D IDFLOLW\ WR FUHDWH DQG PDLQWDLQ D GLVWULEXWHG DQG G\QDPLF UHVHUYDWLRQ VWDWH RYHU D PXOWLFDVW RU XQLFDVW WUDQVPLVVLRQ SDWK 5693 DOVR SURYLGHV PDQ\ RWKHU LPSRUWDQW IHDWXUHV IRU VLJQDOOLQJ LQ WKH ,Qn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

PAGE 102

)LJXUH 5693 )XQFWLRQLQJ %ORFN 'LDJUDPV VLQJOH QHWZRUN OLQN HDFK UHFHLYHU KDV QR QHHG WR PDNH GLVWLQFW UHVHUYDWLRQV 5693 UHDOL]HV WKHVH HIILFLHQFLHV ZKHQ LQVWDOOLQJ UHVHUYDWLRQV LQ URXWHUV E\ PHUJLQJ UHVHUYDn WLRQ UHTXHVWV DORQJ WKH PXOWLFDVW GLVWULEXWLRQ WUHH WR WKH VRXUFH $V VKRZQ LQ )LJ WKH UHVHUYDWLRQ IRU D VLQJOH UHFHLYHU GRHV QRW QHHG WR WUDYHO WR WKH VRXUFH RI D PXOWLFDVW WUHH UDWKHU WKH UHVHUYDWLRQ WUDYHOV RQO\ XQWLO LW UHDFKHV D UHVHUYH EUDQFK RI WKH WUHH 2QH FDQ HDVLO\ VKRZ WKDW WKH UHVHUYDWLRQ RYHUKHDG LV UHGXFHG IURP 1f WR 1ORJ1ff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

PAGE 103

6HQGHU 5HFHLYHU 5HFHLYHU 5HFHLYHU )LJXUH 5HVHUYDWLRQ 0HUJH RI 5693 $70 DOORZHG LQ D VLQJOH SRLQW WR D PXOWLSRLQW 9LUWXDO &KDQQHO 9&f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f > @ 7KH 573 LV D SURWRFRO SURYLGLQJ VXSSRUW IRU DSSOLFDWLRQV ZLWK UHDOWLPH SURSn HUWLHV LQFOXGLQJ WLPLQJ UHFRQVWUXFWLRQ ORVV GHWHFWLRQ DV ZHOO DV VHFXULW\ 573 KHOSV HQVXUH UHOLDEOH UHDOWLPH GHOLYHU\ E\ DGGLQJ ILHOGV FRQWDLQLQJ WLPHVWDPS DQG VHTXHQFH

PAGE 104

QXPEHU DQG WKHVH ILHOGV FDQ EH XVHG E\ D GHVWLQDWLRQ KRVW WR UHFRQVWUXFW WKH WHPSRn UDO LQIRUPDWLRQ RI 573 SDFNHW VWUHDPV ,Q DGGLWLRQ WKH GHOLYHU\ RU GDWD WUDQVSRUW LV HQKDQFHG E\ DQRWKHU SURWRFRO 573 &RQWURO 3URWRFRO 57&3f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f DQG $70 )RUXP ,3 0XOWLFDVW 3URWRFRO DQG 0XOWLFDVW %DFNERQH ,3 0XOWLFDVW 3URWRFRO IDFLOLWDWHV GLVWULEXWHG DSSOLFDWLRQV WR DFKLHYH WLPH FULWLFDO FRPPXQLFDWLRQV VXFK DV YLGHRFRQIHUHQFLQJ RYHU ZLGH DUHD ,3 QHWZRUNV > @ 7KH ,3 0XOWLFDVW URXWHUV W\SLFDOO\ UHIHUUHG WR DV fPURXWHUVf WDNH WKH UHVSRQVLELOLW\ RI GLVWULEXWLQJ DQG UHSOLFDWLQJ PXOWLFDVW GDWD VWUHDPV WR WKHLU GHVWLQDWLRQV DV RSSRVHG

PAGE 105

)LJXUH 0XOWLFDW %DFNERQH 0%RQHf WR LQGLYLGXDO ,3 KRVWV 7KH 0XOWLFDVW %DFNERQH 0%RQHf WRSRORJ\ RI PURXWHUV LV GHn VLJQHG LQ VXFK D PDQQHU WKDW LW IDFLOLWDWHV fHIILFLHQWf GLVWULEXWLRQ RI SDFNHWV ZLWKRXW RYHUORDGLQJ DQ\ QRGH LQDSSURSULDWHO\ 0%RQH KDV JURZQ WUHPHQGRXVO\ VLQFH LWV LQFHSWLRQ LQ 0%RQH LV HVn VHQWLDOO\ D YLUWXDO QHWZRUN ZKRVH PXOWLFDVW VHUYLFHV KDYH EHHQ FXUUHQWO\ HQMR\HG E\ WKRXVDQGV RI ,QWHUQHW XVHUV ZKR SDUWLFLSDWH LQ FRQIHUHQFHV LQ D ORFDO RU JOREDO OHYHO +RZHYHU 0%RQH LV QR ORQJHU D VLPSOH YLUWXDO QHWZRUN VLWWLQJ RQ WRS RI WKH ,QWHUQHW EXW LV UDSLGO\ EHLQJ LQWHJUDWHG LQWR WKH ,QWHUQHW LWVHOI 0%RQH LV FDSDEOH RI VXSn SRUWLQJ ,3 PXOWLFDVW DQG WKH QRGHV RI 0%RQH DUH OLQNHG E\ YLUWXDO SRLQWWRSRLQW OLQNV FDOOHG fWXQQHOVf DV VKRZQ LQ )LJXUH 7KH WXQQHO HQGSRLQWV DUH W\SLFDOO\ ZRUNVWDWLRQFODVV KRVWV KDYLQJ RSHUDWLQJ V\VWHP VXSSRUW IRU ,3 PXOWLFDVW DQG UXQn QLQJ WKH fPURXWHGf PXOWLFDVW URXWLQJ GDHPRQ 7KHVH KRVWV DFW DV URXWHUV RI PXOWLn FDVW E\ XVLQJ D URXWLQJ SURWRFRO FDOOHG 'LVWDQFH 9HFWRU 0XOWLFDVW 5RXWLQJ 3URWRFRO '9053f >@ 6LQFH WKH ILUVW DXGLR FRQIHUHQFH LQ 0%RQH KDV VHHQ WKH GHYHORSPHQW RI D ZLGH UDQJH RI QHZ DSSOLFDWLRQV XVLQJ DXGLR YLGHR ZKLWHERDUG DQG WH[W DV PHGLD 0%RQH VHVVLRQV DUH D FRPELQDWLRQ RI WKHVH PHGLXP ,QIRUPDWLRQ DERXW WKHVH

PAGE 106

VHVVLRQV LV SHULRGLFDOO\ WUDQVPLWWHG DFURVV WKH 0%RQH RQ D ZHOONQRZQ PXOWLFDVW DGGUHVV 8VLQJ WKH 6HVVLRQ 'L5HFWRU\ WRRO 6'5f DV VKRZQ LQ )LJXUH 0%RQH XVHUV FDQ REWDLQ D OLVW RI DGYHUWLVHG VHVVLRQV IURP ZKLFK RQH FDQ ODXQFK WKH 0%RQH WRROV UHTXLUHG WR MRLQ WKH VHVVLRQV DV ZHOO DV FUHDWH D QHZ VHVVLRQ ([DPSOHV RI WKH 0%RQH WRROV IRU YLGHR VWUHDP9,& DQG DXGLR VWUHDP9$7 DUH VKRZQ LQ )LJXUH 7R FUHDWH D PXOWLFDVW VHVVLRQ DV VKRZQ LQ )LJXUH RQH PXVW VSHFLI\ YDOLG &ODVV ,3 DGGUHVVHV IRU LWV SURJUDP HOHPHQWV DXGLR YLGHR ZKLWHERDUG DQG WH[Wf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n PHQW RI DQDO\VLV VWXG\ LQ WKH IROORZLQJ VHVVLRQ 7UDIILF 0HDVXUHPHQWV DQG $QDO\VLV 1HWZRUN WRSRORJ\ DQG 1HWZRUN WRRO 7KH PHDVXUHPHQW DQG DQDO\VLV IURP QHWZRUN WUDFHV KDV EHHQ ZLGHO\ YLHZHG DV RQH RI WKH PRVW HIILFLHQW ZD\V WR VWXG\ QHWZRUN G\QDPLFV XVDJH FKDUDFWHULVWLFV DQG JURZWK SDWWHUQV 7KH PHDVXUHPHQW LQ WKLV VHVVLRQ ZDV FRQGXFWHG LQ WKH /DERn UDWRU\ IRU ,QIRUPDWLRQ 6\VWHPV DQG 7HOHFRPPXQLFDWLRQV /,67f DW WKH 8QLYHUVLW\ RI

PAGE 107

)LJXUH 6HVVLRQ GLUHFWRU\ 6'5f )ORULGD $XJXVW 7KH QHWZRUN WRSRORJ\ RI WKH /,67B1HW LV RXWOLQHG LQ )LJXUH /,67B1HW LV D %DVH7 (WKHUQHW FRPSULVHG RI WZR 6XQ6SDUF :RUNVWDWLRQV DQG PRUH WKDQ WHQ 3HQWLXP GHVNWRS FRPSXWHUV 7KH WZR :RUNVWDWLRQV fFRPODEOf DQG fUREXVWf DUH UXQQLQJ 6XQ26 DQG 6RODULV UHVSHFWLYHO\ ZKHUHDV WKH GHVNWRS FRPSXWHUV DUH UXQQLQJ D IXOO UDQJH RI RSHUDWLRQ V\VWHPV LQFOXGLQJ :LQGRZV 17 :LQGRZV :RUNVWDWLRQ :LQGRZV DQG /LQX[ $ fPURXWHf GDHPRQ LV UXQQLQJ LQ fUREXVWf WR HQDEOH WKH WUDQVPLVVLRQ DQG UHFHSWLRQ RI 0%RQH VHVVLRQV $ VRIWZDUH f,&$67f GHYHORSHG E\ WKH ,&$67 &RUSRUDWLRQ LV LQVWDOOHG LQ fJDQGDOIf fOLVWOf DQG fOLVWf WR IDFLOLWDWH WKH UHFHSWLRQV RI 0%RQH VHVVLRQV RQ KRVWV UXQQLQJ :LQGRZV RU :LQGRZV 17 :H XVHG D QHWZRUN WRRO fWFSGXPSf IRU GDWD FROOHFWLRQV LQ WKH /,67B1HW 7FSGXPS LV D WRRO IRU QHWZRUN PRQLWRULQJ DQG GDWD DFTXLVLWLRQ DQG ZDV RULJLQDOO\ GHYHORSHG E\ 9DQ -DFREVRQ DW /DZUHQFH %HUNHOH\ 1DWLRQDO /DERUDWRU\ >@ DV SDUW RI DQ RQJRLQJ UHVHDUFK SURMHFW WR LQYHVWLJDWH DQG LPSURYH 7&3 DQG LQWHUQHW JDWHZD\

PAGE 108

,7=6HVVLRQ 1DPH_/,67 WHVW UXQ 'HVFULSWLRQ 6GU &UHDWH 1HZ 6HVVLRQ 7KLV LV D PXOWLFDVW WHVW UXQ LQ WKH /DE IRU ,QIRUPDWLRQ 6\VWHPV DQG 7HOHFRPPXQLFDWLRQV LW \ 85/ ZZZOLVWXIOHGX 7HVW 85/ (QFU\SWLRQ 7\SH RI 6HVVLRQ 7HVW 6FRSH 0HFKDQLVP A 77/ 6FRSH Y $GPLQ 6FRSH 6FRSH A 6LWH Y 5HJLRQ Y :RUOG 0HGLD ?I? IL DXGLR 3URWRFRO 573 )RUPDW 3&0 YLGHR 573 + ;_ ZKLWHERDUG ;, WH[W $GGUHVV 3RUW cc 6HVVLRQ ZLOO EH DFWLYH 2QFH IQRP_7XH 1RY DW c IRU KRXUV IURP M IURP  DW n IRU DW IRU 5HSHDW IRU 3HUVRQ WR FRQWDFW DERXW WKLV VHVVLRQ 2O_:HQ
PAGE 109

Df YLGHR Ef DXGLR I7O /,67 WHVW UXQ B Â’ :HQ
PAGE 110

)LJXUH /,671(7 7RSRORJ\ SHUIRUPDQFH 2YHU WKH SDVW IHZ \HDUV WKLV WRRO KDV EHHQ VWHDGLO\ LPSURYHG E\ WKH H[FHOOHQW FRQWULEXWLRQV IURP WKH ,QWHUQHW FRPPXQLW\ :LWK 7FSGXPS RQH FDQ SULQW RXW WKH SDFNHWV RQ D QHWZRUN LQWHUIDFH LQ YDULRXV H[SUHVVLRQV DFFRUGLQJ WR SURWRFROV SRUW QXPEHUV SDFNHW OHQJWK DQG VR RQ 7KH WLPH UHVROXWLRQ RI 7FSGXPS LV VHF 7UDIILF 0HDVXUHPHQWV DQG $QDO\VLV :H PRQLWRU DQG FROOHFW WKH SDFNHW IORZV LQ WKH /,67B1HW E\ UXQQLQJ 7FSGXPS LQ D /LQX[ KRVW fODWFKPDQf 7KH SDFNHWV DUH FROOHFWHG ILUVW DQG WKHQ ILOWHUHG RXW DFFRUGLQJ WR WKH DOJRULWKP GHILQHG LQ D VKHOO VFULSW 7KH WUDFHV UHSUHVHQW ERWK /$1 DQG :$1 WUDIILF LQ D VHQVH WKDW WKH :$1 WUDIILF LV FRUUHVSRQGLQJ WR WKRVH SDFNHWV GHVWLQHG WR RU RULJLQDWHG IURP H[WHUQDO QHWZRUNV ZKLOH WKH /$1 WUDIILF DUH WKRVH RULJLQDWHG IURP DQG GHVWLQHG WR WKH /,67B1HW 7KH RULJLQDO WUDFH LV IXUWKHU GLYLGHG LQWR VHYHUDO JURXSV DFFRUGLQJ WR WKH SURWRFRO ZLWK ZKLFK HDFK SDFNHW LV DVVRFLDWHG 7KH SURWRFROV RI LQWHUHVW DUH f,3 0XOWLFDVWf 0%RQHf f7UDQVSRUW &RQWURO 3URWRFRO 7&3ff DQG f8VHU 'DWDJUDP 3URWRFRO 8'3ff

PAGE 111

)LJXUH 7UDFHV LQ /$1V DQG :$1V )LJXUH VKRZV DQ H[DPSOH RI D PLQXWH WUDFH ZKLFK ZDV FROOHFWHG LQ WKH /,67 IURP 30 WR 30 $XJXVW ,Q WKLV )LJXUH f0%RQHf UHSUHVHQWV WKH ,3 0XOWLFDVW SDFNHWV JHQHUDWHG IURP 0%RQH VHVVLRQV f7&3f UHSUHVHQWV 7&3 SDFNHWV LQ DSSOLFDWLRQV VXFK DV ::: +773f 7HOQHW IWS DQG HPDLO 6073f DQG f8'3f UHSUHVHQWV 8'3 SDFNHWV LQ DSSOLFDWLRQV VXFK DV 9,& 9$7 6'5 DQG &X6HH0( :H QRWLFHG WKDW ERWK 7&3 DQG 8'3 WUDFHV UHYHDO EXUVW\ DUULYDO UDWHV ZKLOH WKH 0%RQH WUDFH UHPDLQV DW D VWHDG\ OHYHO WKURXJK WKH HQG RI WKH UXQ RI 7FSGXPS 7R FRPSDUH WKH WUDIILF FKDUDFWHULVWLFV RI GLIIHUHQW SURWRFROV WKH DUULYDO UDWH VWDWLVWLFV DUH VXPPDUL]HG LQ 7DEOH :H XVH 9DULDQFHWLPH 3ORWV 973f WR HVWLn PDWH WKH +XUVW SDUDPHWHU DV VKRZQ LQ )LJXUH 1RWLFH WKDW 0%RQH WUDIILF KDV DQ HVWLPDWHG +XUVW SDUDPHWHU HTXDO WR ZKLFK LPSOLHV 6KRUW 5DQJH 'HSHQGHQFH 65'f ZKLOH ERWK 7&3 DQG 8'3 WUDIILF KDV RQH LQ WKH UDQJH RI > @ LPSO\LQJ

PAGE 112

3URWRFRO S SDFNHWVHFf D SDFNHWVHFf + 0%21( 7&3 8'3 $// 7DEOH $UULYDO UDWH VWDWLVWLFV p f R F &2 &2 2 &' 1 • ( 2f B 2 7 ,7 S r0%21( 2 7&3 8'3 [ $// L ; 4 Q ; Y U + ; [ A n X R ; [ + % R r R + A + + r ORJ$JJUHJDWLRQ OHYHO Pff )LJXUH 9DULDQFH WLPH SORW /RQJ 5DQJH 'HSHQGHQFH /5'f ,Q DGGLWLRQ WKH HVWLPDWHG +XUVW SDUDPHWHU IRU WKH RYHUDOO WUDIILF LV ZKLFK LPSOLHV /5' DV ZHOO )LJXUH )LJXUH DQG )LJXUH VKRZ WKH 4XDQWLOH4XDQWLOH SORW 44SORWf IRU 0%RQH 7&3 DQG 8'3 WUDIILF UHVSHFWLYHO\ 7KH SXUSRVH RI WKH 44 SORWV KHUH LV WR GHWHUPLQH ZKHWKHU WKH WUDFHV FRPH IURP D 3RLVVRQ GLVWULEXWLRQ UHn JDUGOHVV RI SDUDPHWHU YDOXHV ,I WKH WUDFH GRHV FRPH IURP D 3RLVVRQ GLVWULEXWLRQ WKH SORW ZLOO EH OLQHDU 2WKHUZLVH D FXUYDWXUH ZLOO EH LQWURGXFHG LQ WKH SORW ,W LV HYLGHQW WKDW ERWK WKH 7&3 DQG 8'3 WUDFHV GHYLDWH UDGLFDOO\ IURP D 3RLVVRQ GLVWULEXWLRQ ZKLOH WKH 0%RQH WUDFH LV PDWFKHG E\ D 3RLVVRQ GLVWULEXWLRQ TXLWH ZHOO )LJXUH

PAGE 113

)LJXUH 44SORW IRU 0ERQH WUDIILF GHSLFWV WKH 44SORW IRU WKH RYHUDOO WUDIILF DQG FRQILUPV WKDW D 3RLVVRQ WUDIILF PRGHO LV LQDSSURSULDWH IRU DJJUHJDWHG /$1 DQG :$1 WUDIILF &RQYHQWLRQDO WUDIILF PRGHOV DV ZHOO DV RWKHU WUDIILF PRGHOV VXLWDEOH IRU WUDIILF GLVSOD\LQJ VHOIVLPLODULW\ DQG /5' DUH GLVFXVVHG LQ &KDSWHU

PAGE 114

)LJXUH 44SORW IRU 7&3 WUDIILF )LJXUH 44SORW IRU 8'3 WUDIILF

PAGE 115

)LJXUH 44SORW IRU RYHUDOO WUDIILF

PAGE 116

&+$37(5 &21&/86,21 $70 1HWZRUNV DQG 6HOIVLPLODULW\ $V\QFKURQRXV 7UDQVIHU 0RGH $70f KDV EHHQ ZLGHO\ FRQVLGHUHG WR EH RQH RI WKH PRVW SURPLVLQJ VROXWLRQV IRU WKH QH[W JHQHUDWLRQ KLJK VSHHG DQG PXOWLPHGLD QHWZRUNV 7KH DQWLFLSDWLRQ LV PDLQO\ GXH WR WKH VFDODEOH EXLOGLQJ EORFNV LW SURYLGHV WR VHUYH DV D VWDEOH IRXQGDWLRQ IRU HYROYLQJ SXEOLF DQG SULYDWH QHWZRUNV $70 LV GHVLJQHG WR KDQGOH PDQ\ NLQGV RI LQIRUPDWLRQ LQFOXGLQJ YRLFH GDWD LPDJH DQG YLGHR LQ DQ LQWHJUDWHG PDQQHU DQG LV H[SHFWHG WR VXSSRUW DSSOLFDWLRQV ZLWK GLYHUVH WUDIILF FKDUDFWHULVWLFV DQG TXDOLW\ RI VHUYLFH 4R6f UHTXLUHPHQWV 7KH GLVFRYHU\ RI VHOIVLPLODU EHKDYLRUV LQ /RFDO $UHD 1HWZRUNV /$1Vf DQG :LGH $UHD 1HWZRUNV :$1Vf LV RQH WKH PRVW UHPDUNDEOH PLOHVWRQHV LQ WKH VWXG\ RI FRPPXQLFDWLRQ QHWZRUNV RI ODWH 7R IDFLOLWDWH WKH GHVLJQ DQG RSWLPDO SHUIRUPDQFH RI IXQFWLRQDO $70 QHWZRUNV WZR NH\ LVVXHV DUH DGGUHVVHG LQ WKLV 3K' UHVHDUFK QDPHO\ DQ DSSURSULDWH FKDUDFWHUL]DWLRQ DQG PRGHOLQJ RI $70 WUDIILF DQG HIIHFWLYH FRQJHVWLRQ PDQDJHPHQW VFKHPHV VXLWDEOH IRU WKH VHOIVLPLODU RU IUDFWDOOLNH WUDIILF H[n SHFWHG RQ $70 QHWZRUNV 7R VWXG\ QHWZRUN G\QDPLFV DQG UHDO WUDIILF FKDUDFWHULVWLFV ZH FRQGXFWHG WUDIILF PHDVXUHPHQWV DQG DQDO\VLV EDVHG RQ HPSLULFDO WUDFHV FROOHFWHG LQ WKH /DERUDWRU\ IRU WKH ,QIRUPDWLRQ 6\VWHPV DQG 7HOHFRPPXQLFDWLRQV /,67f :H IRXQG WKDW WKH WUDFHV GLVSOD\HG VHOIVLPLODU FKDUDFWHULVWLFV ,Q DGGLWLRQ WKH VLPXODn WLRQ UHVXOWV LQ RXU UHVHDUFK FRQILUPHG WKDW WKH WUDIILF LQ /$1:$1V DQG WKH WUDIILF JHQHUDWHG E\ 9%5 VHUYLFH LQ $70 QHWZRUNV DUH PRUH DSSURSULDWHO\ PRGHOHG E\ SURn FHVVHV EDVHG RQ VHOIVLPLODU LQVWHDG RI WUDGLWLRQDO 3RLVVRQEDVHG SURFHVVHV ,Q RWKHU

PAGE 117

ZRUGV WKH IORZ FRQWURO VFKHPHV ZKLFK XVH 3RLVVRQEDVHG WUDIILF PRGHOV UHVXOWHG LQ DQ RYHUHVWLPDWLRQ RI WKHLU SHUIRUPDQFH 3UHYHQWLYH )ORZ &RQWURO 3UHYHQWLYH IORZ FRQWURO JHQHUDOO\ LQYROYHV WKH IROORZLQJ WZR SURFHGXUHV FDOO DGPLVVLRQ FRQWURO &$&f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n ZRUNV ZH SURSRVHG D SUHYHQWLYH IORZ FRQWURO VFKHPH XVLQJ D &RQQHFWLRQ $GPLVVLRQ &RQWURO &$&f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n ZRUN GHVLJQHUV WKHUHIRUH IDFH D FKDOOHQJH 7KH EXUVW\ DQG F\FOLF QDWXUH RI 9DULDEOH %LW 5DWH 9%5f WUDIILF FUHDWHV DQRWKHU LVVXH IRU WUDQVPLVVLRQ LQ $70 QHWZRUNV 7R

PAGE 118

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f FRQWUROOHU DQG D OLQHDU SUHGLFWRU WR DGDSWLYHO\ XSGDWH WKH RSWLPXP GDWD HPLVVLRQ UDWH DW WKH WUDQVPLWWHU 7KH DGDSWLYH SROLF\ DWWHPSWV WR NHHS WKH EXIIHU RFFXSDQF\ IRU HDFK YLUWXDO FKDQQHO DW D VWHDG\ OHYHO DQG WKH VLPXODWLRQ UHVXOWV VKRZHG WKDW WKH SURSRVHG VFKHPH ZRUNHG HIIHFWLYHO\ DJDLQVW QHWZRUN FRQJHVWLRQ ,Q DGGLWLRQ WR WKH GHVLJQ RI WKH QHZ IORZ FRQWURO VFKHPH ZH DOVR GHYHORSHG D KLHUDUFKLFDOO\ VWUXFWXUHG WHVWEHG WR PHDVXUH QHWZRUN SHUIRUPDQFH DQG H[SORUH YDULRXV IORZ FRQWURO VFKHPHV LQ $70 QHWZRUNV ZLWK GLYHUVH FODVVHV RI LQFRPLQJ WUDIILF )XWXUH :RUN )XWXUH GLUHFWLRQV UHJDUGLQJ WKLV ZRUN DUH VXPPDUL]HG DV IROORZV f 7UDIILF FKDUDFWHUL]DWLRQ 7KH WUDFHV FROOHFWHG LQ WKH /,67 UHSUHVHQW W\SLFDO QHWZRUN EHKDYLRUV LQ D VPDOO ZRUNJURXS HQYLURQPHQW 7KH VWXG\ RI WUDIILF FKDUDFWHULVWLFV LQ D ODUJHU HQYLURQn PHQW VXFK DV 'HSDUWPHQW RU &DPSXV ZLOO EH EHQHILFLDO WR WKH GHVLJQ RI $70 VZLWFKHV DQG IORZ FRQWURO VFKHPHV LQ WKH EDFNERQH RI ZLGH DUHD QHWZRUNV f 3UHYHQWLYH IORZ FRQWURO VFKHPH ,Q WKH SURSRVHG &$& HDFK VRXUFH KDV WR VXEPLW WKH WUDIILF SDUDPHWHUV LQFOXGLQJ + D + DQG GHVLUHG &HOO /RVV 5DWLR H 7KLV &$& LV WKHUHIRUH EHWWHU VXLWHG IRU DSSOLFDWLRQV VXFK DV 9LGHR RQ 'HPDQG 92'f 2Q WKH RWKHU KDQG IRU WKH &$& WR EH DEOH WR VXSSRUW UHDOWLPH WUDIILF WKHVH SDUDPHWHUV PXVW EH FRPSXWHG

PAGE 119

RQ WKH IO\ ,W SRVHV D FKDOOHQJH HVSHFLDOO\ IRU HVWLPDWLQJ WKH +XUVW SDUDPHWHU ZKLFK UHTXLUHV ORQJHU WLPH LQWHUYDO WR REWDLQ PRUH DFFXUDWH HVWLPDWLRQ f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f DQG SUHGLFWRUV EDVHG RQ $5 PRGHO RU IX]]\ ORJLF FDQ EH XVHG DW WKH H[SHQVH RI FRPSOH[LW\ 2I SDUWLFXODU QRWH LV WKH IDFW WKDW ERWK HIIHFWLYHQHVV DQG VLPSOLFLW\ DUH NH\ UHTXLUHPHQWV RI D IORZ FRQWURO VFKHPH VR WKH WUDGHRIIV DUH ZRUWK IXUWKHU LQYHVWLJDWLRQ

PAGE 120

$33(1',; $ 6&5,376 )25 *(1(5$7,1* 6(/)6,0,/$5 75$)),& $O 5DQGRP 0LGSRLQW 'LVSODFHPHQW $OJRULWKP 5DQGRP 0LGSLQW 'LVSODFHPHQW $OJRULWKP b [ LQWHQGHG +XUVW SDUDPHWHU \ QXPEHU RI SRLQW WR JHQHUDWH IXQFWLRQ >DEWVFDOH@ EURZQ[\f K [ PD[OHYHO IORRUORJ\fORJff VWDUW O OHQJWK aPD[OHYHOO W OVTUWf§frKff D ]HURVOHQJWKf E ]HURVOHQJWKf VFDOH ]HURVPD[OHYHOf fVFDOH IDFWRU IRU L OPD[OHYHO VFDOHLf WraKrLf§ff HQG fVXEGLYLVLQ DHQJWKf UDQGQ EOHQJWKf DOHQJWKf OHYHO O

PAGE 121

,OO ZKLOH OHYHO PD[OHYHO IRU L OaOHYHOf VWDUW OaPD[OHYHOOHYHOOfrLOf ILQLVK VWDUWaPD[OHYHOOHYHOf PLG IORRUILQLVKVWDUWff EPLGf UDQGQ WHPS EPLGfrVFDOHOHYHOf DPLGf DVWDUWfDILQLVKffWHPS HQG OHYHO OHYHOO HQG b HQG RI 50' $OJRULWKP $ 3RZHU 6SHFWUXP 'HQVLW\ $SSUR[LPDWLRQ r )DVW $SSUR[LPDWLRQ RI 6HOIVLPLODU VDPSOH XVLQJ ))7 f GHYHORS E\ 9HUQ 3D[VRQ b K LQWHQGHG +XUVW SDUDPHWHU n Q QXPEHU RI SRLQW WR EH JHQHUDWHG IXQFWLRQ >[@ VVSKQf Q Q ODPGD OQfrSLQ SVG SVGBIJQKODPGDf U UDQGOQf UH[S ORJOUf ISVG SVGrUH[SQf SKDVH rSLrUDQGOQf ] VTUWSVGfrFRVSKDVHfMrVLQSKDVHff ]Qf DEV]Qff

PAGE 122

]PLUU FRQMIOLSOU]ff ]H[SDQG > ] ]PLUUQf@ [ UHDOLIIW]H[SDQGff SORW[f UD IXQFWLRQ >SVG@ SVGBIJQKODPGDf Q VL]HODPGDf G rKO GSULPH rK $ rVLQSLrKfrJDPPDGfrRQHVO QfFRVODPGDff DO rSLrRQHVQfODPGD EO rSLrRQHVOQfODPGD D rrSLrRQHVQfODPGD E rrSLrRQHV QfODPGD D rrSLrRQHVQfODPGD E rrSLrRQHVQfODPGD D rrSLrRQHVQfODPGD E rrSLrRQHVQfODPGD % DOaGEOaGDaGEaGDfGEaG DaGSULPH EaGSULPH DaGSULPH EaGSULPHfSLK SVG $r>DEVODPGDffaG%@ rf HQG RI 36' DSSUR[LPDWLRQ

PAGE 123

$33(1',; % 6&5,376 )25 (67,0$7,1* +8567 3$5$0(7(5 %O 9DULDQFH 7LPH 3ORW $OJRULWKP r/ +XUVW SDUDPHWHU (VWLPDWRU E\ 973 b [ VHOIVLPLDOU VHTXHQFH r K WDUJHW KXVUW SDUDPHWHU IXQFWLRQ >YWS@ YWSK[f Q VL]H[f YDULOf FRY[f U LQF IRU L LQFLQFU P URXQGaLf QD IORRUQPf [D UHVKDSH[OPrQDfPQDf [D VXP[DfP YDULLLQFOf FRY[Df HQG YWS ORJO2YDULYDULOff ILJXUHf KROG RQ SORWLQFU YWS fRff KROG RQ SORWU rKfrUff

PAGE 124

RKROG RQ D[LV> U @f WLWOHf9DULDQFH7LPH 3ORWff \ODEHOfORJO2 1RUPDOL]HG 9DULDQFHfff [ODEH.fORJO2 $JJUHJDWLRQ /HYHO Pf fff b HQG RI 973 $OJRULWKP % 5HVFDOHG $GMXVWPHQW $OJRULWKP +XUVW SDUDPHWHU (VWLPDWRU E\ 56 [ VHOIVLPLDOU VHTXHQFH K WDUJHW KXVUW SDUDPHWHU IXQFWLRQ >UV Q@ UV[Kf QD PD[VL]H[ff V ]HURVf QL LQF M L Vf ZKLOH VMfQD M ML VMf OMOfrQL HQG FRXQW MO UV ]HURVFRXQW f Z ]HURVFRXQW f IRU L OFRXQW IRU N VLfQD ZNf VXP[ONfffNrPHDQ[OVLfff HQG

PAGE 125

UVLf PD[PD[ZffPLQPLQZfffYDU[VLfQDff HQG UV ORJUVf Q ORJVFRXQWff SORWQUVfff HQG RI 56 $OJRULWKP

PAGE 126

5()(5(1&(6 >@ : /HODQG 0 7DTTX : :LOOLQJHU DQG :LOVRQ 2Q WKH VHOIVLPLODU QDWXUH RI (WKHUQHW WUDIILF H[WHQGHG YHUVLRQf ,((($&0 7UDQVDFWLRQV RQ 1HWZRUNLQJ SDJHV >@ 0 : *DUUHWW DQG : :LOOLQJHU $QDO\VLV PRGHOLQJ DQG JHQHUDWLRQ RI VHOIn VLPLODU 9%5 YLGHR WUDIILF ,Q 6,*&200 6\PSRVLXP RQ &RPPXQLFDWLRQV $Un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f 6DQ )UDQVLVFR &DOLIRUQLD 0DUFK >@ 5 3D]K\DQQXU DQG 5 $JUDZDO $QDO\WLFDO DQG QXPHULFDO UHVXOWV IRU IHHGEDFN EDVHG IORZ FRQWURO RI %,6'1$70 QHWZRUNV ZLWK VLJQLILFDQW SURSDJDWLRQ GHn OD\V ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf %RVWRQ 0DVVDFKXVHWWV $SULO >@ +XQW &UHGLWEDVHG )&9& SURSRVDO IRU $70 WUDIILF PDQDJHPHQW UHYLVLRQ 5O 7HFKQLFDO 5HSRUW 5 $70 )RUXP 0D\ >@ $ 'DLOLDQDV DQG $ %RYRSRXORV 5HDOWLPH DGPLVVLRQ FRQWURO DOJRULWKPV ZLWK GHOD\ DQG ORVV JXDUDQWHHV LQ $70 QHWZRUNV ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf %RVWRQ 0DVVDFKXVHWWV $SULO >@ 6 .LP DQG % /HH 7KUHHOHYHO WUDIILF VKDSHU DQG LWV DSSOLFDWLRQ WR VRXUFH FORFN IUHTXHQF\ UHFRYHU\ IRU 9%5 YLGHR VHUYLFHV LQ $70 QHWZRUNV 7UDQVDFWLRQV RQ 1HWZRUNLQJ f $XJXVW >@ 6 < /X ,QWHJUDWHG 7UDIILF &RQWURO 0HFKDQLVLPV IRU $70 1HWZRUNV 3K' WKHVLV 8QLYHUVLW\ RI )ORULGD >@ 6 < /X DQG + $ /DWFKPDQ $QDO\VLV DQG RSWLPL]DWLRQ RI D SDUWLDO EXIIHU VKDULQJ VFKHPH IRU $70 VZLWFK RYHUORDG FRQWURO 3URFHHGLQJV RI WKH 0XOWLPHGLD &RPSXWLQJ &RQIHUHQFH 6DQ -RVH SDJHV

PAGE 127

>@ 0 5LWWHU 1HWZRUN EXIIHU UHTXLUHPHQWV RI WKH UDWHEDVHG FRQWURO PHFKDQLVP IRU $%5 VHUYLFHV ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf 6DQ )UDQVLVFR &DOLIRUQLD 0DUFK >@ 6 (O+HQDRXL 5 &RHOKR DQG 6 7RKPH $ EDQGZLGWK DOORFDWLRQ SURWRFRO IRU 03(* 9%5 WUDIILF LQ $70 QHWZRUNV ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf 6DQ )UDQVLVFR &DOLIRUQLD 0DUFK >@ + $ EDWFK PDQ DQG : < )X 3HUIRUPDQFH RI $70 QHWZRUNV XVLQJ VHOIVLPLODU YV 3RLVVRQ WUDIILF PRGHOV ,Q 3URFHHGLQJV RI WKH ,$67(' ,QWHUQDWLRQDO &RQn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n FHPEHU >@ + )RZOHU DQG : ( /HODQG /RFDO DUHD QHWZRUN WUDIILF FKDUDFWHULVWLFV ZLWK LPSOLFDWLRQV IRU EURDGEDQG QHWZRUN FRQJHVWLRQ PDQDJHPHQW ,((( -RXUQDO RQ 6HOHFWHG $UHDV LQ &RPPXQLFDWLRQV >@ 5 *XVHOOD &KDUDFWHUL]LQJ WKH YDULDELOLW\ RI DUULYDO SURFHVVHV ZLWK LQGLFHV RI GLVSHUVLRQ ,((( MRXUQDO RQ 6HOHFWHG $UHDV LQ FRPPXQLFDWLRQV >@ 6 $ +HLPOLFK 7UDIILF FKDUDFWHUL]DWLRQ RI WKH 1)61(7 QDWLRQDO EDFNERQH 3URFHHGLQJ RI WKH :LQWHU 8VHQL[ FRQIHUHQFH SDJHV -DQXDU\ >@ : ( /HODQG 0 6 7DTT : :LOOLQJHU DQG 9 :LOVRQ 2Q WKH VHOIVLPLODU QDWXUH RI (WKHUQHW WUDIILF ,Q '3 6LGKX HGLWRU 6,*&200 6\PSRVLXP RQ &RPPXQLFDWLRQV $UFKLWHFWXUHV DQG 3URWRFROV SDJHV 6DQ )UDQFLVFR &DOn LIRUQLD 6HSWHPEHU $&0 DOVR LQ &RPSXWHU &RPPXQLFDWLRQ 5HYLHZ f 2FW >@ 0 +HOOVWHUP DQG 3 ( :LUWK 7UDIILF PRGHOV IRU ,6'1 GDWD XVHUV 2IILFH DXWRPDWLRQ DSSOLFDWLRQ 7HOHWUDIILF DQG 'DWDWUDIILF LQ D 3HULRG RI &KDQJ $ -HQVHQ 9 % ,YHUVHQ (GVf QRUWK +ROODQG SDJHV

PAGE 128

>@ 5 $GGLH 0 =XNHUPDQ DQG 7 1HDPH )UDFWDO WUDIILF 0HDVXUHPHQWV PRGn HOLQJ DQG SHUIRUPDQFH HYDOXDWLRQ ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf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n QLWH YDULDQFHV ,QWHUQDWLRQDO (FRQRPLFDO 5HYLHZ SDJHV >@ & +XDQJ 0 'HYHWVLNLRWLV /DPEDGDULV DQG $ 5 .D\H 6HOIVLPLODU PRGHOn LQJ RI YDULDEOH ELW UDWH FRPSUHVVHV YLGHR $ XQLILHG DSSURDFK $&0 6,*&200 >@ & : *UDQJHU DQG 5 -R\HX[ $Q LQWURGXFWLRQ WR ORQJPHPRU\ WLPH VHULHV PRGHOV DQG IUDFWLRQDO GLIIHUHQFLQJ 7LPHV VHULHV $QDO >@ 9 3D[VRQ )DVW DSSUR[LPDWLRQ RI VHOIVLPLODU QHWZRUN WUDIILF 7HFKQLFDO 5HSRUW /%/ /DZUHQFH %HUNHOH\ /DERUDWRU\ DQG ((&6 'LYLVLRQ 8QLYHUVLW\ RI &DOLIRUQLD %HUNHOH\ &DOLIRUQLD $SULO >@ : /HODQG 0 7DTTX : :LOOLQJHU DQG :LOVRQ 6HOIVLPLODULW\ LQ KLJKn VSHHG SDFNHW WUDIILF $QDO\VLV DQG PRGHOLQJ RI (WKHUQHW WUDIILF PHDVXUHPHQWV 6WDWLVWLFDO 6FLHQFH >@ 0 %HUQVWHLQ DQG + 1JX\HQ %ORFNLQJ UHGXFWLRQ LQ IDVW UHVHUYDWLRQ SURn WRFROV ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf 7RURQWR &DQDGD -XQH >@ 6 / %HUPHMR DQG + 3HWLW %DQGZLGWK UHVRXUFH GLPHQVLRQLQJ LQ $70 QHWn ZRUNV $ WKHRUHWLFDO DSSURDFK DQG VRPH VWXG\ FDVHV ,Q WK ,QWHUQDWLRQDO 7HOHWUDIILF &RQJUHVV YROXPH SDJHV &RSHQKDJHQ -XQH >@ .DQJ < @ +DUPV DQG 4 +RQJ $ FDOO DGPLVVLRQ VFKHPH IRU $70 QHWZRUNV EDVHG RQ UHILQHPHQW RI WUDIILF PHDVXUHV ,Q &RQIHUHQFH 5HFRUG RI WKH ,QWHUQDWLRQDO &RQIHUHQFH RQ &RPPXQLFDWLRQV ,&&f YROXPH SDJHV

PAGE 129

>@ & &KDQJ DQG 5 &KHQJ 7UDIILF FRQWURO LQ DQ $70 QHWZRUN XVLQJ IX]]\ VHW WKHRU\ ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf 7RURQWR &DQDGD -XQH >@ 6 0DQWKRUSH DQG ; *DUFLD 6KDSHU GLPHQVLRQLQJ IRU 0$1 WR $70 EDVHG :$1 ,Q ,QWHURS (QJLQHHUVf &RQIHUHQFH /DV 9HJDV &DOLIRUQLD 0D\ >@ GH 9HFLDQD DQG :DOUDQG 7UDIILF VKDSLQJ IRU $70 QHWZRUNV 'UDIW 'HSDUWn PHQW RI (OHFWULFDO (QJLQHHULQJ DQG &RPSXWHU 6FLHQFHV 8QLYHUVLW\ RI &DOLIRUQLD %HUNHOH\ &DOLIRUQLD 'HFHPEHU >@ 3 6NHOO\ 3DFLILFL DQG 0 6FKZDUW] $ FHOO DQG EXUVW OHYHO FRQWURO IUDPHZRUN IRU LQWHJUDWHG YLGHR DQG LPDJH WUDIILF ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf 7RURQWR &DQDGD -XQH >@ $ 5 5HLEPDQ DQG $ : %HUJHU 7UDIILF GHVFULSWRUV IRU 9%5 YLGHR WHOHFRQIHUn HQFLQJ RYHU $70 QHWZRUNV ,((($&0 7UDQVDFWLRQV RQ 1HWZRUNLQJ f -XQH >@ $ $WDL DQG +XL 5DWHEDVHG IHHGEDFN WUDIILF FRQWUROOHU IRU $70 QHWZRUNV ,Q &RQIHUHQFH 5HFRUG RI WKH ,QWHUQDWLRQDO &RQIHUHQFH RQ &RPPXQLFDWLRQV ,&&f YROXPH SDJHV 1HZ 2UOHDQV /RXLVLDQD 0D\ ,((( >@ 0 'HFLQD 7 7RQLDWWL 3 9DFFDUL DQG / 9HUUL %DQGZLGWK DVVLJQPHQW DQG YLUWXDO FDOO EORFNLQJV LQ $70 QHWZRUNV ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf SDJHV 6DQ )UDQFLVFR &DOn LIRUQLD -XQH ,((( >@ + (VDNL ,ZDPXUD DQG 7 .RGDPD $ VLPSOH DQG HIIHFWLYH DGPLVVLRQ FRQWURO PHWKRG IRU DQ $70 QHWZRUN ,Q 3URFHHGLQJV RI WKH ,((( &RQIHUHQFH RQ *OREDO &RPPXQLFDWLRQV */2%(&20f SDJHV 6DQ 'LHJR &DOLIRUQLD 'HFHPEHU ,((( >@ 3 %R\HU $ FRQJHVWLRQ FRQWURO IRU WKH $70 ,Q ,QWHUQDWLRQDO 7HOHWUDIILF &RQJUHVV 6HYHQWK 6SHFLDOLVW 6HPLQDU SDJH 0RUULVWRZQ 1HZ -HUVH\ 2FWRn EHU ,7& >@ + 7 .XQJ 5 0RUULV 7 &KDUXKDV DQG /LQ 8VH RI OLQNE\OLQN IORZ FRQWURO LQ PD[LPL]LQJ $70 QHWZRUN SHUIRUPDQFH VLPXODWLRQ UHVXOWV ,Q $SSHDUHG LQ 3URF ,((( +RW ,QWHUFRQQHFWV 6\PSRVLXP f SDJH SDJHV $XJXVW >@ 1RUURV $ VWRUDJH PRGHO ZLWK VHOIVLPLODU LQSXW 4XHXHLQJ 6\VWHPV SDJHV >@ 5 *XHULQ + $KPDGL DQG 0 1DJKVKLQHK (TXLYDOHQW FDSDFLW\ DQG LWV DSSOLFDn WLRQ WR EDQGZLGWK DOORFDWLRQ LQ KLJKVSHHG QHWZRUNV ,((( -RXUQDO RQ 6HOHFWHG $UHDV LQ &RPPXQLFDWLRQV >@ :DOUDQG $Q ,QWURGXFWLRQ WR 4XHXHLQJ 1HWZURNV 3UHQWLFH+DOO >@ 9 ( %HQHV *HQHUDO 6WRFKDVWLF 3URFHVVHV LQ WKH 7KHRU\ RI 4XHXHV $GGLVRQ :HVOH\ >@ 1 'XIILHOG DQG 1 2f&RQQHOO /DUJH GHYLDWLRQV DQG RYHUIORZ SUREDELOLWLHV IRU WKH JHQHUDO VLQJOHVHUYHU TXHXH ZLWK DSSOLFDWLRQV 3URF &DPEULGJH 3KLO 6RF

PAGE 130

>@ / 5REHUWV &DQ $%5 VHUYLFH UHSODFH 9%5 VHUYLFH LQ $70 QHWZRUNV ,Q &203&21 f 7HFKQRORJLHV IRU WKH ,QIRUPDWLRQ 6XSHUKLJKZD\ 'LJHVW RI 3Dn SHUV SDJHV >@ 3 3DQFKD DQG 0 ( =DUNL 03(* FRGLQJ IRU YDULDEOH ELW UDWH YLGHR WUDQVPLVn VLRQ ,((( &RPPXQLFDWLRQV 0DJD]LQH SDJHV PD\ >@ : 'LQJ -RLQW HQFRGHU DQG FKDQQHO UDWH FRQWURO RI 9%5 YLGHR RYHU $70 QHWZRUNV ,Q ,((( 7UDQVDFWLRQV RQ &LUFXLWV DQG 6\VWHPV IRU 9LGHR 7HFKQRORJ\ SDJHV >@ & % /HH % +D DQG 5 + 3DUN &RPSXWDWLRQ RI HIIHFWLYH EDQGZLGWK RI DJJUHJDWHG 9%5 03(* YLGHR WUDIILF LQ $70 QHWZRUNV XVLQJ WKH PRGLILHG HTXLYDOHQW FDSDFLW\ ,Q &RQIHUHQFH 5HFRUG RI WKH ,QWHUQDWLRQDO &RQIHUHQFH RQ &RPPXQLFDWLRQV ,&&f YROXPH SDJHV >@ 3 *R\DO + 0 9LQ & 6KHQ DQG 3 6KHQR\ $ UHOLDEOH DGDSWLYH QHWZRUN SURWRFRO IRU YLGHR WUDQVSRUW ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDWLRQV ,((( ,QIRFRPf YROXPH SDJHV >@ / & +X DQG /LQ $ VLPXODWLRQ VWXG\ RI UHDOWLPH 03(* WUDIILF XVLQJ IRUZDUG HUURU FRQWURO VFKHPH REMHFWRULHQWHG UHDOWLPH GHSHQGDEOH V\VWHPV ,Q 3URFHHGLQJV RI :25'6f SDJHV >@ 0 5 ,VPDLO ( /DPEDGDULV 0 'HYHWVLNLRWLV DQG $ 5 .D\H 0RGHOOLQJ SULn RULWL]HG 03(* YLGHR XVLQJ 7(6 DQG D IUDPH VSUHDGLQJ VWUDWHJ\ IRU WUDQVPLVVLRQ LQ $70 QHWZRUNV ,Q 3URFHHGLQJV RI WKH &RQIHUHQFH RQ &RPSXWHU &RPPXQLFDn WLRQV ,((( ,QIRFRPf SDJHV >@ & 6 /L < 2IHN DQG 0 @ 5 %UDGHQ / =KDQJ DQG HW DO 5HVRXUFH UHVHUYDWLRQ SURWRFRO YHUVLRQ IXQFn WLRQDO VSHFLILFDWLRQ 7HFKQLFDO UHSRUW ,(7) 'HFHPEHU >@ 5 )UHGHULFN + 6FKXO]ULQQH 6&DVQHU DQG 9 -DFREVRQ 5WS $ WUDQVSRUW SURn WRFRO IRU UHDOWLPH DSSOLFDWLRQV ,Q 5)& QXPEHU 5)& ,(7) >@ + 6FKXO]ULQQH 5WS SURILOH IRU DXGLR DQG YLGHR FRQIHUHQFHV ZLWK PLQLPDO FRQWURO ,Q 5)& QXPEHU 5)& ,(7) >@ +RIIPDQ DQG )HUQDQGR 5WS SD\ORDG IRUPDW IRU PSHJPSHJ YLGHR ,Q 5)& QXPEHU 5)& ,(7) >@ & $OPHURWK DQG 0 + $PPDU 0XOWLFDVW JURXS EHKDYLRU LQ WKH LQWHUQHWfV PXOWLFDVW EDFNERQH PERQHf ,Q ,((( &RPPXQLFDWLRQV 0DJD]LQH SDJHV -XQH >@ 0 5 0DFHGRQLD DQG 3 %UXW]PDQ 0ERQH SURYLGHV DXGLR DQG YLGHR DFURVV WKH ,QWHUQHW ,Q &RPSXWHU SDJHV $SULO >@ 6 ( 'HHULQJ DQG 5 &KHULWRQ 0XOWLFDVW URXWLQJ LQ GDWDJUDP LQWHUQHWZRUNV DQG H[WHQGHG /$1V ,Q $&0 7UDQVDFWLRQV RQ &RPSXWHU 6\VWHPV SDJHV 0D\ >@ 1HWZRUN 5HVHDUFK *URXS /DZUDQFH %HUNHOH\ 1DWLRQDO /DERUDWRU\ 7FSGXPS

PAGE 131

%,2*5$3+,&$/ 6.(7&+ :HQ
PAGE 132

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

PAGE 133

7KLV GLVVHUWDWLRQ ZDV VXEPLWWHG WR WKH *UDGXDWH )DFXOW\ RI WKH &ROOHJH RI (QJLQHHULQJ DQG WR WKH *UDGXDWH 6FKRRO DQG ZDV DFFHSWHG DV SDUWLDO IXOILOOPHQW RI WKH UHTXLUHPHQWV IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 0D\ :LQIUHG 0 3KLOOLSV 'HDQ &ROOHJH RI (QJLQHHULQJ .DUHQ $ +ROEURRN 'HDQ *UDGXDWH 6FKRRO


LIST OF TABLES
2.1 Technology Comparison 7
2.2 Service Characteristics of AAL Layers 14
2.3 QoS Parameters 15
4.1 Performance comparison of Poisson process and SSP traffic models . 55
5.1 Robustness of the proposed CAC scheme 69
5.2 Guarantee of QoS 71
6.1 CLR and normalized utilization rate under various (K/, K^) 85
6.2 CLR and normalized utilization rate vs. number of VBR VCs .... 89
7.1 Arrival rate statistics 102
vi


Ill
while level <= maxlevel,
for i=l:2~(level-1),
start=l+2~(maxlevel-level+l)*(i-l);
finish=start+2~(maxlevel-level+1);
mid = floor((finish+start)/2.0);
b(mid)=randn;
temp=b(mid)*scale(level);
a(mid)=(a(start)+a(finish))/2.0+temp;
end
level=level+l;
end
% end of RMD Algorithm
A.2 Power Spectrum Density Approximation
*/. Fast Approximation of Self-similar sample using FFT
/. develop by Vern Paxson
% h: intended Hurst parameter
'/. n: number of point to be generated
function [x]=ssp(h,n)
n=n/2;
lamda=(l:n)*pi/n;
psd=psd_fgn(h,lamda);
r=rand(l,n);
rexp=-log(l-r);
fpsd=psd.*rexp(1,n);
phase=2*pi*rand(l,n);
z=sqrt(psd).*(cos(phase)+j*sin(phase));
z(n)=abs(z(n));


37
where a is a finite positive constant independent of m, and (3 = 2 2H. On the other
hand, for short-range dependent processes for which A771'^ (k) = 0,as m OO, it
can be shown as well that
var(X<'m')) > bm~l, as m oo, (3.22)
where b is a finite positive constant independent of m. The Hurst parameter estima
tion based on (3.21) is called Variance-time plot (vtp). A Matlab script for vtp is
included in Appendix A.2.
3.3 Modeling of Self-Similarity
For decades, self-similarity has been a phenomenon that has been studied ex
tensively in several other fields including hydrology [31] and economics [35]. Because
of this exhaustive research, several formal mathematical models have already been
developed which exhibit self-similarity. However, due to the abstract nature of these
models, a practical interpretation for the self-similarity is difficult to find. Two such
models are presented in this section: the fractional Gaussian noise (fGn), and the
fractional autoregressive integrated moving average models (f-ARIMA). We also pre
sented fraction Brownian motion (fBm), the companion of fGn, and a construction
based on aggregating many simple renewal rewards processes. The latter is of par
ticular significance to this research as it provides a practical interpretation for the
self-similarity found in Ethernet LAN traffic.
3.3.1 Fractional Brownian Motion
The ordinary Brownian motion, B(t), describes the movement of a particle in
a liquid, subjected to collision and other forces. This motion is a self-similar process
with Hurst parameter 1/2 and independent Gaussian increments such that,
B(t)=|t|1/2B(1)
(3.23)


3
1.2 Dissertation Outline
This dissertation is organized as follows: In Chapter 2, we give an introduction
of ATM technology along with various traffic characteristics in ATM networks. We
describe the ATM cell structure in great detail. The bottom three layers of B-ISDN
reference model, namely Physical Layer, ATM Transport Layer and ATM Adaptation
Layer are presented. Traffic characteristics of Constant Bit-rate (CBR), Variable Bit-
rate (VBR) and Available Bit-rate (ABR) or Best-effort services and QoS in ATM
networks are discussed at the end of this chapter.
In Chapter 3, we first review traditional traffic models including Poisson,
Markov and Markov-Modulated Poisson model. We then focus on self-similar pro
cesses and their properties and implication for novel self-similar traffic models. Two of
the most important properties of self-similar processes, namely, Hurst effect and long
range dependence will be discussed and analyzed completely. We introduce variance
time plots (VTP) and rescaled adjusted (R/S) statistics to estimate the Hurst pa
rameter of a self-similar sample path. We present several synthesized algorithms
for generating self-similar samples, for example, Random Mid-point Displacement
(RMD) algorithm and FFT approximation. In this research, most algorithms for es
timating the Hurst parameter and generating self-similar sample path are converted
in Matlab scripts.
In Chapter 4, we introduce a traffic control and management framework in
ATM networks. Important parts of the framework are Connection Admission Control
(CAC), Resource Management (RM), Traffic Shaping, Feedback Control, and so on.
We describe several examples including rate-basted and credit-based flow control
schemes. The performance of several congestion control schemes are re-evaluated by
using self-similar traffic models.


97
Figure 7.4: Session directory (SDR)
Florida, August 1997. The network topology of the LIST_Net is outlined in Figure
7.8. LIST_Net is a 10/100 Base-T Ethernet comprised of two SunSparc Workstations
and more than ten Pentium desktop computers. The two Workstations, comlabl
and robust, are running SunOS 4.1.3 and Solaris 2.4, respectively, whereas the
desktop computers are running a full range of operation systems including Windows
NT 4.0, Windows Workstation 4.0, Windows 95, and Linux. A mroute daemon
is running in robust to enable the transmission and reception of MBone sessions.
A software ICAST developed by the ICAST Corporation is installed in gandalf,
listl, and list4 to facilitate the receptions of MBone sessions on hosts running
Windows 95 or Windows NT.
We used a network tool, tcpdump, for data collections in the LIST_Net.
Tcpdump is a tool for network monitoring and data acquisition and was originally
developed by Van Jacobson at Lawrence Berkeley National Laboratory [74] as part
of an ongoing research project to investigate and improve TCP and internet gateway


38
and
E[(B(tx + t) B(ti))5(*i)] = 0. (3.24)
From (3.23) and (3.24), it follows that
E{[B(tx + r) B^)}2} = |r| E[B2{ 1)] (3.25)
Mandelbrot [31] defines fractional Brownian motion (fBm) with Hnrst param
eter H as being the moving average of dB(t) in which past increments of B(t) are
weighted by the kernel (t s)H~1'2,
= mrw) /> s)"-1/2 +jf(* -
(3.26)
where T denotes the gamma function: T(r + 1) = r-r(r) and T(k + 1) = k\ as k being
an integer. By definition in (3.4), we have
BH(t)=tHBH( 1). (3.27)
Several important properties of fBm are as follows.
1. Bh(0) = 0
2. E{BH(t)} = 0
3. i + r) BH(tx))2} = \t\2H E{B2H{ 1)}
4. E{BH{tx)BH(t2)} = \(\tx\2H + \h\2H \tx t2\2H) E{B2h( 1)}
3.3.2 Fractional Gaussian Noise
The increments of fBm, G(n), form a stationary sequence called fractional
Gaussian noise (fGn).


APPENDIX B
SCRIPTS FOR ESTIMATING HURST PARAMETER
B.l Variance Time Plot Algorithm
L Hurst parameter Estimator by VTP
% x: self-simialr sequence
/, h: target husrt parameter
function [vtp]=vtp(h,x)
n=size(x,2);
vari(l)=cov(x);
r=3;
inc=.141421;
for i=inc:inc:r,
m=round(10~i);
na=floor(n/m);
xa=reshape(x(l:m*na),m,na);
xa=sum(xa)/m;
vari(i/inc+l)=cov(xa);
end
vtp=loglO(vari./vari(l)) ;
figure(2);
hold on
plot(0:inc:r, vtp, o);
hold on
plot(0:r, -(2-2*h)*(0:r));
113


27
3.1 Poisson-Based Traffic Models
3.1.1 Poisson Process Models
Poisson process models are the oldest traffic models, dating back to the ad
vent of telephony primarily contributed by the well-known pioneering engineer A. K.
Erlang in the 1910s. It is well known that the traffic on telecommunication networks
can be modelled and characterized by a Poisson process. One of the most important
traffic characteristics on telecommunication networks is that most human initiated
processes are memoryless. A Poisson process can be characterized as a renewal pro
cess whose interarrival times {/} are exponentially distributed with rate of A, that
is
Pr{/ < t} = 1 exp(Ai). (3.1)
Equivalently, it is a counting process, satisfying
Pr{A(i) n) = p exp(At), (3.2)
where A[t) is the number of arrivals in a time interval (0, t).
Poisson processes enjoy some elegant analytical properties such as:
1. The number of arrivals in disjoint intervals are independent, which is referred
to as independent increment property.
2. Merging independent Poisson processes result in a new Poisson process with a
rate being the sum of the component rates.
The independent increment property renders Poisson a memoryless process.
This, in Pirn, greatly simplifies queueing problems involving Poisson arrivals. In
spite of the analytic simplification, Poisson processes do have a significant modeling


41
a fractional ARIMA(p, d, q) process satisfies equations Eq. 3.6, which makes it an
asymptotical self-similar processes with LRD and Hurst parameter H d + 1/2.
Fractional ARIMA(p, d, q) processes were first introduced in [37]. One advan
tage of fractional ARIMA(p, d, q) processes over fractional Gaussian noise is their
flexibility with respect to modeling both short-range and long-range dependence,
making them better suited to modeling phenomenon such as VBR video traffic [36].
3.3.4 Self-Similarity Through Aggregation
Consider a number of independent sources which alternate between the states
on and off (renewal-rewards processes) where the amount of time spent in each
state is randomly distributed with a heavy-tailed distribution. An example of such
a heavy-tailed distribution is the stable Pareto distribution (see Appendix B of [5])
with parameter 1 < a < 2. If we construct a new sequence consisting of the number
of sources which are on at points in time separated by a fixed length interval then it
has been shown in [1] that the resulting sequence will be asymptotically self-similar.
The notion of generating a self-similar process by aggregating a number of
simple renewal rewards processes provides intuition into the self-similarity of Ethernet
traffic as well as into the nature of a single source on an Ethernet network. If
we consider each workstation (or more likely each user) attached to an Ethernet
network as a renewal-rewards process where the process is on if the workstation is
communicating over the network and off otherwise, then we see that the amount
of traffic on the network can be thought of as the aggregation of a number of renewal
rewards processes. This also leads to the conclusion that the communication bursts
produced by individual users have a heavy-tailed distribution. This conclusion is
supported by in depth analysis of Ethernet [1] and wide area network traffic traces
[38].


74
6.1 Introduction
Asynchronous Transfer Mode (ATM) is widely considered to be the next gen
eration of high speed internetworking technology mainly due to the scalable building
blocks it provides to serve as a stable foundation for evolving public and private
networks. With the advent of Broadband ISDN, ATM networks must support appli
cations with diverse traffic characteristics and quality of service (QoS) requirements.
Therefore, an accurate traffic characterization is of primary importance in implement
ing effective flow control strategies and making efficient use of network resources in
ATM networks.
Network traffic is usually characterized by the tolerance of information delay
or loss. For example, the traffic for Available Bit Rate (ABR), or best-effort service,
demands superior Cell Loss Rate (CLR) but yields inferior performance in terms of
Cell Transfer Delay (CTD) and Cell Delay Variation (CDV). On the other hand, real
time traffic such as that for Constant Bit Rate (CBR) service is distinguished by its
stringent CTD and CDV requirements. Real-time traffic contrasts significantly from
elastic or best-effort traffic by primarily supporting applications that have dedicated
bandwidth. These applications, however, do have some ability to adapt to network
conditions such as cell delay or cell losses but only within a margin of tolerance [59].
The other type of real-time traffic in addition to CBR traffic is Variable Bit Rate
(VBR) traffic such as that generated in videoconferencing using compression. The
characteristics of VBR video traffic depends primarily on two factors:
1. the nature of the video, namely whether there are frequent or rare scene changes;
2. the coding scheme employed such as, for example, MPEG, H.261 etc.
For teleconferencing and videotelephone applications, the CCITT H.261 stan
dard specifies compression techniques at rates of px64 kilobits/second where p ranges
from 1 to about 30. Another well-known compression algorithms for video is MPEG,


54
Figure 4.2: Variance-time plot for synthesized sample paths
verified by the results shown in this Figure [11], the ATM Forum has selected rate-
based flow control scheme to be used for flow and congestion control in ATM networks.
Thus, in studying the effects of self-similar traffic on the network performance, we
concentrate on the BECN scheme with or without the slow start option. Note that
in this session, rate-based scheme I and rate-based scheme II are referred to the
BECN scheme without and with slow-start, respectively.
Figure 4.4 and Figure 4.5 show the goodput of VC-1 for the BECN scheme
with and without slow-start, respectively. In both figures, Poisson process as well as
SSP with H=0.7 and 0.9, are used as the traffic models. The values of goodput at
steady state are listed in Table 4.1 where the first column is the traffic model for the
sources, the second and fourth columns are the flow control schemes, and the third
and fifth columns are the percentage of over-estimation of goodput by using Poisson
process instead of SSP traffic models. Evidence shows that using a Poisson process


23
will follow the same route in traversing a switch, the sequence of cells is preserved
without the need of explicit sequence numbers.
The most important function of the egress adaptor is to temporarily buffer
those cells which cannot be delivered to a receiver immediately. A finite buffer mem
ory allocated in the egress adaptor is for resolving output contention. If congestion
causes the buffers to fill up, the controller has no choice but to discard cells. The
buffer thus forms a critical resource that needs to be adequately controlled.
A modified Partial Buffer Sharing scheme is implemented in the egress adap
tor to ensure satisfactory loss rates for different loss priority classes and fairness in
sharing the resource among competing users. Moreover, a priority-dependent ser
vice scheduling scheme is realized to control the order in which cells are sent from
the buffer to the output link. The service scheduling control is necessary for meet
ing the different switching delay requirements for different service priority classes.
Service and loss priorities are assigned to each virtual connection during connection
establishment based on the delay and loss requirements specified for the particular
connection.
It is noticed that only one ingress adaptor can have access to the high-speed
bus at any time. So a bus arbitration scheme is implemented for resolving contention
for access among different ingress adaptors. Since the shared medium operates at a
substantially higher speed (approximately 5 Gbps) than that of link interface mod
ules, the CPS-100 can be viewed as a non-blocking switch, having no internal buffering
and with no contention visible from outside the switching fabric. Though the high
speed of the switching fabric reduces input contention, it also results in major queue
ing taking place at the buffer of egress adaptors. The proposed flow control scheme
can be used along with existing buffer management schemes to avoid immense buffer
overflows.


83
Figure 6.5: Network Scenario
reverse direction. Due to the balanced network topology, similar network behavior
at each CPE is expected from simulations.
6.5 Numerical Results and Discussions
To evaluate the performance of the proposed PID flow control scheme under
a congested network condition, each VC for voice generates traffic contributing 20%
of the bandwidth of the trunk link, while each VC for video is contending for the
To CPS 100-A
From CPS 100-A
VBR
) VC for CBR
) VC for VBR
( VC for Feedback
Figure 6.6: Functional model of CPE


101
Figure 7.9: Traces in LANs and WANs
Figure 7.9 shows an example of a 200-minute trace, which was collected in the
LIST from 1:00 PM to 4:20 PM, August 25 1997. In this Figure, MBone represents
the IP Multicast packets generated from MBone sessions, TCP represents TCP
packets in applications such as WWW (HTTP), Telnet, ftp, and e-mail (SMTP),
and UDP represents UDP packets in applications such as VIC, VAT, SDR, and
Cu-SeeME. We noticed that both TCP and UDP traces reveal bursty arrival rates,
while the MBone trace remains at a steady level through the end of the run of
Tcpdump.
To compare the traffic characteristics of different protocols, the arrival rate
statistics are summarized in Table 7.1. We use Variance-time Plots (VTP) to esti
mate the Hurst parameter, as shown in Figure 7.10. Notice that MBone traffic has
an estimated Hurst parameter equal to 0.5 which implies Short Range Dependence
(SRD), while both TCP and UDP traffic has one in the range of [0.75, 0.90] implying


103
Figure 7.11: QQ-plot for Mbone traffic
depicts the QQ-plot for the overall traffic and confirms that a Poisson traffic model
is inappropriate for aggregated LAN and WAN traffic. Conventional traffic models
as well as other traffic models suitable for traffic displaying self-similarity and LRD
are discussed in Chapter 3.


2
Recent studies based on the trace of local area network (LAN) traffic, mainly
at Bellcore [1], have led to the conclusion that LAN traffic can not be adequately
represented by traditional Markov-based models, but instead can be more appropri
ately matched by self-similar or fractal-like models. More recently, variable-bit-rate
(VBR) video traffic [2, 3, 4] and wide area network (WAN) traffic [5] was also found
to exhibit self-similar characteristics. With the foreseeable prevalence of Internet
services and multimedia applications, ATM networks will be carrying a substantial
portion of self-similar traffic.
As we know when a number of bursty traffic sources are active and no interac
tive or preventive action is taken to against it, cell losses tend to occur due to buffer
overflows. Despite that researchers have proposed many flow and congestion control
schemes to solve the problems, whether these approaches take advantage of tech
niques in feedback control [6, 7, 8], call admission control (CAC) [9], traffic shaping
and smoothing [10], or resource reservation and buffer management [11, 12, 13, 14],
none of them takes the self-similarity of the traffic into consideration. As a conse
quence, a new flow control mechanism incorporating more realistic traffic models is
necessary to avoid the tending degradation [15] of network performance and, at the
same time to maintain better network utilization.
The main objective of this research is to design an effective flow and congestion
control scheme for ATM networks by developing a queueing model to gain better
understandings of the behaviors of a network whose incoming traffic is self-similar in
nature. The primary goal of the flow and congestion control mechanism is to minimize
the impact of traffic overload which induces congestion as well as cell loss, and,
simultaneously, to achieve a better network utilization. In addition, this research will
also undertake a tremendous efforts to develop a hierarchically structured testbed for
exploring network performance and various flow control schemes for ATM networks
under mixed self-similar and Poisson traffic.


67
(5.25)
(5.26)
Given p(k) = £, the bandwidth C required to accept a connection can be
computed from (5.26) and is
C = p + H(2pv\ne)1^2H\ ^ V1 l/H\
1 H
(5.27)
5.5 Proposed Call Admission Control Algorithm
The simplest Connection Admission Control strategy can be obtained con
sidering Peak Cell Rate (PCR) allocation for each traffic source. PCR can lead to
no loss of cells but to a very poor link utilization. To obtain a better utilization
of resources, it is necessary to accept a small degradation of the Quality of Service
(QoS) while taking into account the statistical properties of actual traffics. To this
end, we propose a Call Admission Control (CAC) algorithm based on (5.27). The
procedure of the proposed CAC algorithm is described below.
1. Each source has to submit the traffic parameters including /i, v, and H, and
desired Cell Loss Ratio e.
2. Compute required bandwidth C according to (5.27).
3. Compare required bandwidth C with available bandwidth C.
4. Accept the connection, if C < C, or reject, otherwise.
5.6 Numerical Results and Discussions
Assume that the offered traffic source has the parameter (u,v) = (1000,100).
To study the influence of buffer size on cell loss probability, we set the allocated
bandwidth C to be 1050 cell/sec. Figure 5.1 depicts the cell loss probability vs.
buffer size for several Hurst parameters, H. It is seen that when H=0.5 or H=0.6 the


24
2.4.2 Advantages and Limitations of CPS-100 Switch
A principal advantage of the CPS-100 ATM switch structure is its flexibil
ity to support multiple kinds of port interfaces. It has been widely recognized that
ATM technology is targeted to support a rich mixture of broadband and narrowband
services as well as the high-capacity interconnection of existing data networking ap
plications. Thus, it is important for a switching node to provide different kinds of
interfaces for the diverse applications in a highly flexible and expandable way. In
the CPS-100 switch, this is achieved through a modular design that places the ports
on separate interface modules. Various combinations of port interfaces, link access
speeds and physical media can be supported easily by implementing different inter
face modules on the switch. A CPS-100 switch can be configured to support up to
16 of these interface modules.
Another advantage of the CPS-100 switch is that with this architecture dy
namic priority functions and multicast operations can be supported flexibly with no
additional hardware complexity. This function is preferable since some applications
such as ATM LANs require the ability to support multiple guaranteed classes of ser
vices and full-bandwidth multicasting. Each interface module can implement suitable
buffer management and scheduling policies based on the traffic characteristic as well
as loading conditions to best serve the delay and loss requirements of different traffic
classes. In addition, since the major queueing takes place at the output buffer, each
interface module can perform load monitoring on the buffer usages of different traffic
classes to support congestion control operation. Lastly, the buffer space within an
interface module can be allocated dynamically to each port so as to achieve a lower
probability of memory overflow.
On the other hand, as with other shared medium/output buffering ATM
switches, the CPS-100 switch architecture has the typical limitation of requiring the


92
Figure 7.1: RSVP Functioning Block Diagrams
single network link, each receiver has no need to make distinct reservations. RSVP
realizes these efficiencies when installing reservations in routers by merging reserva
tion requests along the multicast distribution tree to the source. As shown in Fig.
7.2, the reservation for a single receiver does not need to travel to the source of a
multicast tree; rather the reservation travels only until it reaches a reserve branch of
the tree. One can easily show that the reservation overhead is reduced from 0(N2) to
0(N-log(N)), where N is the number of receivers. The reduction becomes significant
when the number of receivers is enormous.
Many of the features of RSVP can be supported at the ATM signalling layer
or by the IP over ATM layer. Local policy at the ingress router will determine which
portions of ATM signalling and which of IP over ATM signalling will be used. It is
desirable to modify ATM signalling for QoS heterogeneity and for QoS renegotiation.
The efforts for modifying ATM signalling to better support RSVP can lead to easier
integration of ATM networks into the Internet. The motivation of ATM signalling
modifications falls into two main areas. First, there is currently no mechanism in


34
3.2.2 Self-similarity and Long-Range Dependence
A process satisfying (3.6) is said to exhibit long-range dependence (LRD).
Such processes are characterized by an autocorrelation function r(r) which decays
hyperbolically as r increases. This implication shows that r(r) is non-summable, i.e.
So r(r) = 00 which contrasts with many conventional short-range dependent (SRD)
processes such as Poisson-based processes which have an autocorrelation function
r(r) which decreases exponentially as r increases and is consequently summable,
namely, ^o r(T) < - Furthermore, the non-summability of the correlations cap
tures the intuition behind LRD, namely, that while individual values of r(r) are small
for large r, their cumulative effect is of importance, and produces features which are
drastically different from SRD processes.
Pi'oposition 1 An asymptotically self-similar process X(t) with Hurst parameter H
exhibits LRD as 0.5 < H < 1, and SRD as 0 < H < 0.5.
Proof:
From (3.10), the autocorrelation function of X^ given Hurst parameter H is
r(r) = \[{t + l)2H + (t l)2H 2t2H], We have
r(r) = ir2-2[r2()2 + t2( )2" 2t2]
2 r r
= i'r2ii~2{-r2[(i + r_1)2if + (1 t~1)2H 2]}, (3.12)
(1 + 7--1)2H = 1 + 2Hr-1 + H[2H 1 )r"2 + , (3.13)
and
(1 r_1)2H = 1 2Ht~1 + H(2H 1 )r~2 + . (3.14)
Combining (3.12), (3.13) and (3.14), we obtain


62
where B(t) is a standard Brownian motion, namely a Gaussian process with indepen
dent increments.
(5.2) follows from the equivalent proposition,
^ = *<*> <->
which can be proved by Central Limit Theorem [56].
Extending (5.2) to the case of fGn arrival processes, as suggested in [54],
a(t), the number of arrivals up to time t, can be represented by
a(t) = pt +y/pBH(t), (5.4)
where v is the variance coefficient and Bn(t) is a normalized fractional Brownian
motion (fBm) with zero mean and variance \t\2H. Let 8{n) denote the increment of
A(t) in the interval ((n 1 )T,nT\. We obtain
8(n) = a(nT) a((n 1 )T)
= pT + y/Jw{BH(nT) BH((n 1)T)}
= pT + v/]G(nT), (5.5)
where G(n) is a fractional Gaussian noise with zero mean and a variance equal to
one. From (5.5), it then follows that
E{6(n)} = pT, (5.6)
and
Var(8(n)) = pvE{G2(nT)}
= pv\T\2HE{G2(n)}
= pv\T\2H. (5.7)
From (5.5), (5.6) and (5.7), we conclude that S(n), the increment of A(t), is also
an fGn with Hurst parameter H, mean pT and variance pv\T\2H. From Proposition


53
Figure 4.1: Network scenario for Poisson vs. Self-similar traffic models
goodput, is expected from each of the eight connections. The maximum goodput that
may be achieved in this scenario is 0.125.
The Hurst parameter of the generated sample paths is estimated by Variance
time plot (vtp) as shown in Figure 4.2. Note that the RMD algorithm renders a
Hurst parameter which is slightly smaller than the target value 0.9. In addition, we
noticed that the sample path generated by Poisson process yields a Hurst parameter
0.5 which is quite different from the range of [0.70 0.90] empirically found on real
networks [27]. In our simulations, for both Poisson process and SSP traffic sources,
the egress buffer size is equal to 500 cells and offered traffic load is maintained at the
same level to facilitate meaningful comparisons.
4.3.2 Effective Throughput Comparisons
Figure 4.3 shows the goodput of VC-1, i.e., the virtual connection between Tx-
1 and Rcv-1, for various flow-control schemes where Poisson process traffic models
are used. Despite the obviously superior performance of the credit-based scheme as


82
Rn = Rin {Ki $(e, l) + KD Aen) Rin (6.4)
Clearly we can see that, from (6.4), the instantaneous sending rate of the nth frame
is determined by Kd, a derivative factor, K¡, an integral factor, and l regarding
cumulative error signals. Further, the values and KD control how rapidly the
buffer occupancy approaches the desired buffer size.
6.4 Network Scenario for Simulations
Figure 6.5 shows the network scenario used in our simulations where two
CPS-100 ATM switches, CPS 100-A and CPS 100-B, are connected to each other
with a DS-3 trunk link at a speed of approximately 45 Mbps. The propagation delay
between the two switches is 16.5 ms. There are four CPEs connected to each switch
at both sides of the network. We model the data flows of the network such that they
traverse from sources to corresponding destinations while the feedback transmitted
in the reverse direction. All access ports are at the same speed as that of trunk ports.
We assume a one-to-one correspondence between sources and destinations,
that is, each CPE at the left side of the network makes two VCs, one for voice and
the other for video, to the corresponding CPE at the right. The simplified functional
model of the CPE is shown in Figure 6.6. Due to the stringent QoS requirements
of CBR service, the VCs for voice are given higher service and loss priority than
those of video. Cells from the VCs with same priority will be served on a FIFO
basis. As all traffic aggregates at the outgoing trunk port of Si, the egress buffer
memory and the outgoing link bandwidth of the port become performance bottlenecks
where congestion may occur when too many bursts arrive in a short period. The
resource management (RM) cells carrying buffer information at the egress adapter
are generated periodically and are sent back to all video sources by connections in the


117
[13] M. Ritter. Network buffer requirements of the rate-based control mechanism for
ABR services. In Proceedings of the Conference on Computer Communications
(IEEE Infocom), San Fransisco, California, March 1996.
[14] S. El-Henaoui, R. Coelho, and S. Tohme. A bandwidth allocation protocol for
MPEG VBR traffic in ATM networks. In Proceedings of the Conference on
Computer Communications (IEEE Infocom), San Fransisco, California, March
1996.
[15] H. A. batch man and W. Y. Fu. Performance of ATM networks using self-similar
vs. Poisson traffic models. In Proceedings of the IASTED International Con
ference on Modeling, Simulation and Optimization, Gold Coast, Australia, May
1996.
[16] D. E. McDysan and D. L. Spohn. ATM Theory and Application. Mcgraw Hill,
1994.
[17] Craig Partridge. Gigabit Networking. Addison-Wesley, 1993.
[18] D. Bertsekas and R. Gallager. Data Networks. Prentice Hall, 1992.
[19] D. Benham. A tutorial: ATM in local area networks. Hughes LAN Systems,
1994.
[20] ATM Forum. ATM User-Network Interface Specification. ATM Forum, 1994.
[21] S. Y. Lu, H. A. Latchman, N. Shah W. Y. Fu, A. Khan, and B. Waggener. Shared
midium/output buffering schemes in ATM switching an overview and case
study. International Conference on Computer Communications and Networks,
1994, San Francisco, pages 88-92, 1994.
[22] V. S. Frost and B. Melamed. Traffic modeling for telecommunications networks.
IEEE Communications Magazine, pages 70-81, March 1994.
[23] R. Caceres. Measurements of wide area Internet traffic. Report UCB/CSD
89/550, Computer Sciences Division, University of California, Berkeley, De
cember 1989.
[24] H. J. Fowler and W. E. Leland. Local area network traffic characteristics, with
implications for broadband network congestion management. IEEE Journal on
Selected Areas in Communications, 9:1139-1149, 1991.
[25] R. Gusella. Characterizing the variability of arrival processes with indices of
dispersion. IEEE journal on Selected Areas in communications, 1990.
[26] S. A. Heimlich. Traffic characterization of the NFSNET national backbone.
Proceeding of the 1990 Winter Usenix conference, pages 207-227, January 1990.
[27] W. E. Leland, M. S. Taqq, W. Willinger, and D. V. Wilson. On the self-similar
nature of Ethernet traffic. In D.P. Sidhu, editor, SIGCOMM Symposium on
Communications Architectures and Protocols, pages 183-193, San Francisco, Cal
ifornia, September 1993. ACM. also in Computer Communication Review 23 (4),
Oct. 1992.
[28] K. M. Hellsterm and P. E. Wirth. Traffic models for ISDN data users: Office
automation application. Teletraffic and Datatraffic in a Period of Chang (A.
Jensen, V. B. Iversen Eds) north Holland, pages 162-172, 1991.


36
3.2.3 Self-similarity and Hurst Effect
Self-similar processes also provide an elegant explanation of an empirical law
known as Hursts law or the Hurst effect. For a given set of observations Ad, A2,..., Xn
with sample mean X(n) and sample variance S2(n), the rescaled adjusted range or
11 R/S statistic is given by
max(0, di,d2, dn) min(0, dx, d2,..., dn)
S{n)
where
djt = Xi + X2 + ... + Xk k X(n), k = 1,2,..., n. (3.18)
Hurst [33] found that many naturally occurring time series well represented by the
relation
E[R(n)/S(n)] > a nH, as n * 00 (3.19)
with Hurst parameter H normally around 0.73, and a a finite positive constant in
dependent of n. However, if the observations come from a short-range dependent
(SRD) process, then it has been shown in [31] that
E[R(n)/S(n)] > b n'5, as n > 00 (3.20)
with b another finite positive constant independent of n. A Matlab script for Hurst
parameter estimation based on R/S statistic is included in Appendix A.l.
3.2.4 Self-similarity and Slowly Decaying Variances
Another important feature of self-similar processes is that the variance of
the arithmetic mean, fa, decreases slower than the reciprocal of the sample size m.
Research [34] shows that (3.6) is equivalent to
var(X^) > am~P, as m 00, (3-21)
(3.17)
R(n)/S(n)


66
From (5.13), we have
p{k) = Pr(W(£) > k)
= Pr(max[BH(t) t] > p=z)
>0 V V
= Pr(max[5//(£) ] > 0)
v t>o L w J
Since BH(t)=tHBh( 1), substituting BH(t), we have
P{k)
Pr(max[5H(l)
(C y)t + k
tHV
>0)
Pr(£H(l) > nhn[
(C n)t + k
tHV¡^
Q(min[
^ v n L
(C n)t + k
tHV
]),
where the Q-function is defined by
Q{-X) ~ V2L 6 1 /2dt-
For brevity, denote by f(t). We have
= (1 H){C n)t-H -
Vi
_ tH~\(\-H){C-p)t-kH)
Vi
(5.20)
(5.21)
(5.22)
no =
-H{ 1 H)(C 1 + H{H + l)fc£
i-H-2
V
(5.23)
When t =
kH
= £*, /'(£*) = 0, and
no =
(kH)-H~2[( 1 F)(C A/)p+2(-ff(fcff) + H{H + 1 )k)
V
(kH)-H~2[( 1 H)(C ai)]H+2(Hk)
V
> 0.
(5.24)
Therefore, f(t) has a minimum when t = t*. Substituting t with t*, and using the
approximation Q{x) = e~x2!2 and the theory of large deviations [58], we obtain
(C VitcVfi h)Y~H + k
P(k) = qC- )


115
rs(i)=(max(0,max(w))-min(0,min(w)))/var(x(s(i):na));
end
rs=log(rs);
n=log(s(1:count));
plot(n,rs,+)
1 end of RS Algorithm


56
O 2 4 6 8 10
simulation time (sec)
Figure 4.4: Performance comparison of rate-based scheme I
Figure 4.5: Performance comparison of rate-based scheme II


7
Attributes
Datacom
Telecom
ATM
Traffic Type
data
voice
data, voice, video
Switching
packet
circuit
cell
Unit
variable packet
fixed frame
fixed cell
Quality
best effort
guaranteed
QoS
Access to Media
shared
dedicated
dedicated
Connection
connectionless
connection-oriented
connection-oriented
Table 2.1: Technology Comparison
connections to routers and end systems. This trend has been so marked that it over
takes the development of ATM as a back-bone transmission method and displaces
other development prospects for High-Speed LANs and WANs such as Frame Relay
(FR), Fiber Distributed Digital Interface (FDDI), and Switched Multimegabit Data
Services (SMDS). Additionally, ATM is significantly different from both the popular
data communication and telecommunication technologies in use today. ATM technol
ogy stands out by unifying these two worlds by borrowing desirable attributes from
both data communication and telecommunication technologies [19]. The comparison
between data communication, telecommunication, and ATM technology is shown in
Table 2.1. Explicit information regarding ATM will be discussed in the next few
sections.
2.1.1 ATM Layered Structure
The ATM protocol uses a layered structure, B-ISDN protocol reference model
(see Figure 2.1), similar to the Open System Interconnect (OSI) model. However, the
three-dimensional layer structure of ATM protocol is distinct from all other protocol
models in the use of three planes across all three ATM layers plus a higher layer The
user plane is used for end-to-end data transfer, the control plane supports signalling
used to established a connection, and the management plane supports management


96
sessions is periodically transmitted across the MBone on a well-known multicast
address. Using the Session DiRectory tool (SDR) as shown in Figure 7.4, MBone
users can obtain a list of advertised sessions, from which one can launch the MBone
tools required to join the sessions as well as create a new session. Examples of the
MBone tools for video stream-VIC and audio stream-VAT are shown in Figure 7.6.
To create a multicast session as shown in Figure 7.5, one must specify valid Class D IP
addresses for its program elements (audio, video, whiteboard and text). Valid Class
D IP addresses range from 224.0.0.0 to 239.255.255.255. In addition to IP addresses,
port numbers should be specified for program elements as well. Port numbers are
used by TCP/IP to identify applications. IP multicast packets are encapsulated for
transmission through tunnels, so that they look like normal unicast datagrams to
intervening routers and subnets. A multicast router that wants to send a multicast
packet across a tunnel will attach another IP header in the prefix of the packet,
and set the destination address in the new header to be the unicast address of the
multicast router at the other end of the tunnel. The multicast router at the other end
of the tunnel receives the packet, strips off the encapsulating IP header, and forwards
the packet as appropriate.
Figure 7.7 shows the positions of various network applications in a TCP/IP
layer structure. These applications are of particular interest to our traffic measure
ment of analysis study in the following session.
7.2 Traffic Measurements and Analysis
7.2.1 Network topology and Network tool
The measurement and analysis from network traces has been widely viewed
as one of the most efficient ways to study network dynamics, usage characteristics,
and growth patterns. The measurement in this session was conducted in the Labo
ratory for Information Systems and Telecommunications (LIST) at the University of


63
1 in Chapter 3, <5(n) possesses LRD as H > 1/2. Now let C be the fixed service rate,
i.e., the link speed of an output port, then the number of arrivals up to time t, (3(t),
is equal to Ct. Denoting W(t) the number of cells in the queue up to time t, then by
Reichs Formula [57] we have
W(t) = sup[a(i) ¡3(t) a(x) + f3(x))
x = sup[a(i) a(x) C(t x)]
x = sup [(p-C)(t-x) +y/vp(BH(t)~ BH(x))). (5.8)
x The following propositions are extended from the theorems originally pre
sented in [54] to facilitate the derivation of the cell loss probability for a queueing
system with fGn arrival process.
Proposition 3 W(t) is a self-similar process only if H 1.
Proof:
W(at) sup[(p C) (at ax) + y/vjl(BH(at) Bn(ax))\
x = sup[aH((p C)al~H(t x) + ^/vjl(BH(t) BH(x))\
x = aH sup[a(t) a(x) [p (p C)a}~H](t x)] (5.9)
x As H 1, we obtain
W(at) = asup[o;(t) a(x) C(t x)]
x = alW(t),
which implies that W (t) is a self-similar process with H 1.
Note that, from (5.9), when H ^ 1, W(at), the number of cells in the queue
up to time at has the same distribution as aH times the number of cells in a queue
with the original arrival process but with service rate p (p C)ax~H.


64
Proposition 4 Let p q represent the utilization rate, k the finite buffer size, then
^JfH C'-'/Wk 1+1/H = constant.
Proof:
Since A(t) has stationary increments, W(t) is stationary and thus has the
same distribution as W(0). Substituting t with 0 in (5.8), we obtain
VP(0) = sup[(/i C)(x) + y/vfi(-BH(x)). (5.10)
x<0
Substituting x with -t, and with the fact that Bw(r) is Gaussian, we have
1P(0) = sup[(p-C)(t) + y/vfiBH(t)]
t> o
= may:[y/fwBH{t) (C p)t].
(5.11)
Let p{k) be the cell loss probability for a queue with finite buffer size k, then
we have
p(K) = Pr(W(t) >k) = Pr(bU(0) > k)
= Pr(max[^/JIvBH(t) (C p)t] > k) (5.12)
= Pr(max[B(i) > ^=)- (5-13)
Since Bu(t) is a self-similar process with Hurst parameter H, we have a~H Bh (at)=Bn (t).
Substituting a with bH 1, we obtain
b-lBH(bl'Ht)BH{t)
or, after changing variables,
b-lBH{t)=BH{b~l/Ht).
(5.14)
Let (j) denote Pr(max.[BH(t) at] > b). Using (5.14), we have
(f = Pr(max[H//(i) at] > b)


20
at the University of Florida [21]. The CPS-100 switch1 is an ATM-based switching
system that supports ATM and various data services such as Switch Megabyte Data
Service (SMDS) and Frame Relay for public networks. CPS-100 switch is a time
division, shared medium switch fabric interconnecting time division, shared memory
switching elements. The switch structure is generally composed of the following four
major components:
1. The switching fabric which performs the basic switching function of transferring
cells or packets from input to output. The switching fabric in the CPS-100
switch is composed of a high-speed Virtual ATM (VATM) backplane bus and
is shared with all interface modules.
2. The access interface module provides an interface for customers to access the
services supported by network. The Customer Premises Equipment (CPE)
attaches to an access interface module for transmitting and receiving data via
a dedicated link.
3. The trunk interface module provides an interface for interconnecting switching
systems to support inter-switch communications.
4. The CPU module communicates with all the interface modules in the switch
and performs the high-level control functions, such as connection establishment
and release, memory administration, bandwidth allocation, and maintenance.
2.4.1 Interface Modules
Figure 2.5 illustrates the functional model of a six-port CPS-100 switch. Note
that an access/trunk interface module can support multiple bidirectional links. Both
external trunks and CPE links are connected to the VATM bus via the access/trunk
1The CPS-100 was developed by Loral Data Systems, Sarasota, FL.


87
(a)
(b)
Figure 6.9: Transmitted and received frame size with feedback control


6.1 MPEG frame size and its probability density function 76
6.2 Generic Video Transmission System with Feedback 78
6.3 Generic tracking system and its application in networks 80
6.4 PID flow controller 80
6.5 Network Scenario 83
6.6 Functional model of CPE 83
6.7 Transmitted and received frame size with no feedback control .... 85
6.8 Cumulative number of cells sent and lost with no feedback control . 86
6.9 Transmitted and received frame size with feedback control 87
6.10 Cumulative number of cells sent and lost with feedback control .... 88
7.1 RSVP Functioning Block Diagrams 92
7.2 Reservation Merge of RSVP 93
7.3 Multicat Backbone (MBone) 95
7.4 Session directory (SDR) 97
7.5 Multicast session with audio/video streams 98
7.6 Video and audio streams in a MBone session 99
7.7 Network applications and TCP/IP protocol stack 99
7.8 LIST_NET Topology 100
7.9 Traces in LANs and WANs 101
7.10 Variance -time plot 102
7.11 QQ-plot for Mbone traffic 103
7.12 QQ-plot for TCP traffic 104
7.13 QQ-plot for UDP traffic 104
7.14 QQ-plot for overall traffic 105
viii


10
|4 1 byte ! \* 1 byte !
Figure 2.2: ATM Cell Format at UNI and NNI
be implemented at the UNI, although currently no consensus exists as to how this
will work. This access flow control will use the four Generic Flow Control (GFC) bits
and is only meaningful when there is shared access to one ATM UNI port. At the
receiving end, GFC receives as input a stream of cells from the Physical layer, from
which it performs header extraction and delivers cell, in order, to the appropriate
AAL service access point (SAP) using the Virtual Channel Identifier/Virtual Path
Identifier (VCI/VPI) values as identifiers. At switching elements, the ATM layer uses
the VCI/VPI to route the cells. While the VPI and VCI values may change at each
switching element, the ATM layer does this translation.
Interpretation of the values in the Payload Types (PT) and Cell-Loss Priority
(CLP) fields is done at the ATM layer in switching elements. The main feature of
the PT Identifier is to discriminate user cells that carry user information from non
user cells [20]. The PT Identifier of a user cell may not only indicate whether the cell
has experienced congestion with Explicit Forward Congestion Indication (EFCI), but
also suggest whether it contains an indication to the AAL protocol. For example, the
(b) at NNI
I I I
GFC
1 1 1
VPI
VPI
VCI
VCI
VCI
PT CLP
HEC
PAYLOAD (48 bytes)
(a) at UNI


22
From Link A/B
To Link A
To Link B
Figure 2.6: Adapter Interface of a Shared Medium/Output Buffering ATM Switch


50
The credit-based scheme as described so far is also referred to as Flow Con
trolled Virtual Circuit (FCVC) scheme. This initial static version has two problems.
First, if the credits are lost, the sender will not know it. Second, each VC needs to
reserve the entire round trip worth of buffers even though the link is shared by many
VCs. These problems were solved by introducing a credit resynchronization algorithm
and an adaptive version of the scheme.
The credit resynchronization algorithm consists of both sender and receiver
maintaining counts of cells sent and received for each VC and periodically exchanging
these counts. The difference between the cells sent by the sender and those received
by the receiver represents the number of cells lost on the link. The receiver reissues
that many additional credits for VC.
The adaptive FCVC algorithm consists of giving each VC a fraction of the
round trip delay worth of buffer allocation. The fraction depends on the rate at
which the VC uses the credit. For highly active VC, the fraction is larger, while for
less active, the fraction is smaller. Inactive VCs get a small fixed credit. If a VC
does not use its credits, its observed usage rate over a period is low and the VC
gets smaller buffer allocation (and, hence, credits) in the next period. The adaptive
FCVC reduces the buffer requirements considerably, but also the full capacity of the
link, even if there are no other users.
4.2.2 Rate-Based Scheme
The original proposal for a rate-based approach consists of end-to-end control
using a single-bit to control the network. In the proposal, the switches monitor their
queue lengths and, if congested, set the explicit forward congestion indication (EFCI)
in the cell headers. The destination monitors these indications for a periodic interval
and sends an RM cell back to the source. The sources use an additive increase and
multiplicative decrease algorithm to adjust their rates. This particular algorithm


95
Figure 7.3: Multicat Backbone (MBone)
to individual IP hosts. The Multicast Backbone (MBone) topology of mrouters is de
signed in such a manner that it facilitates efficient distribution of packets without
overloading any node inappropriately.
MBone has grown tremendously since its inception in 1992. MBone is es
sentially a virtual network whose multicast services have been currently enjoyed by
thousands of Internet users who participate in conferences in a local or global level.
However, MBone is no longer a simple virtual network sitting on top of the Internet,
but is rapidly being integrated into the Internet itself. MBone is capable of sup
porting IP multicast, and the nodes of MBone are linked by virtual point-to-point
links called tunnels as shown in Figure 7.3. The tunnel endpoints are typically
workstation-class hosts having operating system support for IP multicast and run
ning the mrouted multicast routing daemon. These hosts act as routers of multi
cast by using a routing protocol called Distance Vector Multicast Routing Protocol
(DVMRP) [73].
Since the first audio conference in 1992 MBone has seen the development
of a wide range of new applications using audio, video, whiteboard, and text as
media. MBone sessions are a combination of these medium. Information about these


15
Attribute
Accuracy
Speed
Cell Error Ratio
X
Cell Loss Rate
X
Cell Misinsertion Rate
X
Cell Transfer Delay
X
Cell Delay Variation
X
Table 2.3: QoS Parameters
unspecified QoS class, no value is specified for the performance parameters, even
though the network may assign those internal objectives. The performance parame
ters as defined in [20] can be further divided in two categories, namely accuracy and
speed, as shown in Table 2.3.
2.2.1 Accuracy
The QoS classes regarding accuracy include Cell Error Ratio (CER), Cell Loss
Rate (CLR) and Cell Misinsertion Rate (CMR). The CER for an ATM connection
is defined as
Crr Number of erroneous cells
Total number of successful transmitted and erroneous cells
where successful and erroneous cells found in severely faulty cells block are excluded.
A cell block is a sequence of cells transmitted consecutively on a connection. A
severely faulty cell block occurs when more than a certain number of error cells, lost
cells or misinserted cells are observed in a received cell block.
The CLR is defined for an ATM connection (VCC or VPC) as
CER Number of lost Cells
Number of transmitted Cells
It is noticed that lost and transmitted cells found in severely faulty blocks should be
excluded from the cell population in computing CLR.
The CMR for an VCC or VPC connection is defined as
^ Number of Misinserted Cells
CMR = : .
I ime interval


51
uses a negative feedback in the sense that RM cells are sent only to reduce the rate
but no RM cells are required to increase the rate. The problem is that RM cells may
be lost due to heavy congestion in feedback, and, consequently, the sources will keep
increasing their sending rate and eventually overload it. This drawback is enhanced
by using a positive feedback where it would require a RM cell to increase the rate
but not on decrease. In a more enhanced version, RM cells are sent for both increase
and decrease.
An example of rate-based flow control scheme using a negative feedback is
Backward Explicit Congestion Notification (BECN) scheme. This method, pre
sented by N.E.T [51, 52], consists of switches monitoring their queue lengths and
sending an BECN cell back to source if congested. The sources reduce their rates by
half on the receipt of the BECN cell until a pre-defined minimum rate is reached. If
no BECN cells are received within a recovery period, the rate for that VC is doubled
once each period until it reaches the peak rate. Note that when the minimum rate is
reached, any further receipt of BECN cell will force the source to stop transmitting.
To achieve fairness, the source recovery period was made proportional to the current
level of transmission rate so that the lower transmission rate, the shorter the source
recovery period [53].
When supporting TCP traffic over ATM networks, a slow-start mechanism
may be used jointly with the BECN scheme to avoid packet losses immediately after
the initial transmission in a overloaded situation. Specifically, instead of starting
the transmission at its peak rate, the source starts with the minimum rate and then
recovers to its peak rate gradually by applying the same recovery procedure as in the
BECN scheme.


19
Figure 2.4: TV News
The fast growth in the number of personal computers and workstations and
the rising need for interconnecting them together with, for example, servers and
printers in an office has led to the fast expansion of LANs. As the need for interof
fice communication rose, LANs were in turn interconnected via bridges and routers.
Consequently, users soon started to demand scalable throughput for large volumes
of data exchange between them. In addition, emerging multimedia applications that
integrate voice, data, image, and video started to change the face of networking in
local and wide areas due to their real time and high bandwidth requirements.
2.4 CPS100 Example of ATM Switching Design
In spite of the strong competition from the evolving Gigabit Ethernet technol
ogy, more and more research and development have been involved with ATM switch
ing technology over the years. The main reason is that ATM switches have proved to
be the best option in the backbone of wide area networks (WANs). In this section,
we present a fast packet switch architecture based on a shared medium and output
buffering scheme for which simulation and performance studies have been conducted


65
= Pr(max[6_1?//() b~xat\ > 1)
= Pr(max[BH{b~l^Ht) b~xat] > 1)
= Pr(max[BH(t) b~1+1^Hat] > 1). (5.15)
Combining (5.13) and (5.15), we obtain
p(k) = Pr(max[B(i) (-C)-+V<£_i)(] > j) (5.16)
= Pr(max[.B//(t) ut] > 1) (5.17)
= /M, (5-18)
where u> = Since Bnif) is Ganssian, from (5.17), it is obvious that
f(u>) is a monotonically decreasing function. Therefore, given a cell loss probability,
say e, there exists an inverse function such that u> = f~l (e) =constant.
Denoting p = we have
UJ
IH Ap
y/f
-1/2 H

Proposition 5 Given a fractional queueing system with MCR p, variance coefficient
v, Hurst parameter H, service rate C and finite buffer size k, the approximation of
the cell loss probability, p(k), is
p(k)= exp(
-(C-p)
2 H
1,22 H
2pv(HH(i Hy-ny
)
(5.19)
Proof:


105
Figure 7.14: QQ-plot for overall traffic


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EZA6R475Z_I3W14G INGEST_TIME 2014-12-06T00:29:20Z PACKAGE AA00024486_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES


70
genuine Hurst parameter
Figure 5.2: Estimated CLP vs. genuine Hurst parameters
the proposed algorithm. The SMG reaches a maximum value 1.92 as H = 0.5. When
H = 0.9, the SMG increase from 1.05 as e = 10-9 to 1.35 as e = 10~3.
We conducted simple simulations to verify the guarantee of QoS, i.e., the cell
loss probability (CLP). The fGn with Hurst parameter 0.9, buffer size 1000 cells, and
(u, v) = (1000,100) is used as the traffic source. The allocated bandwidth is obtained
from (5.27). The simulation results are summarized in Table 5.2 where the first
column is the requested CLP ranging from 10~3 to 10~9, and the second and the third
column show the estimated CLP by simulation for selected CAC schemes. Clearly
the proposed CAC (PCAC) ensures the QoS. However, with the other statistical
allocation algorithm, Equivalent Capacity (EC), the QoS is violated mainly because
of the use of an inappropriate traffic model which fails to capture the characteristics
of self-similarity.


79
skewed by a fixed amount of time from each other. The substreams are trans
mitted with the forward error correction code providing a deterministic bound
on the delay and having the implementation simplicity needed for high-speed
video applications.
Prioritized transmission: Ismail and et al [65] proposed a priority scheme for
MPEG video which is implemented in a software encoder that produces a pro
portionate traffic in both (i.e., high and low) priority partitions for all three
frame types used in MPEG. An ATM multiplexer with a pushout buffer scheme
is implemented to provide priority scheduling at the multiplexer for the two pri
ority partitions.
Traffic shaping and smoothing: Burstiness of encoded traffic may be removed
by smoothing it over an interval. Here one has to make a trade-off between cell
losses due to burstiness and allowed delay. Li and et al [66] suggested a new
flow control function called time-driven priority, which is an internal traffic
shaping mechanism supporting CBR with deterministic guarantees, and VBR
with statistical multiplexing. The mechanism does not require the identification
and separation of packet flows of different real-time sessions/connections inside
the network.
Admission control: This mechanism can be used to accept or reject new con
nections at the connection set-up phase. ATM networks require traffic char
acteristics information such as CLR, PCR, SCR, and CDV for VBR traffic to
guarantee the QoS of a connection.
6.3 Proposed Flow Control Scheme
The proposed flow control scheme is comparable to a classical tracking control
system, in which appropriate feedback is applied to have the system output y[n] track
the input r[n] as Figure 6.3(a) shows. A compensator taking the difference between


REFERENCES
[1] W. Leland, M. Taqqu, W. Willinger, and D. Wilson. On the self-similar nature
of Ethernet traffic (extended version). IEEE/ACM Transactions on Networking,
pages 1-15, 1994.
[2] M. W. Garrett and W. Willinger. Analysis, modeling and generation of self
similar VBR video traffic. In SIGCOMM Symposium on Communications Ar
chitectures and Protocols, pages 269-280, London, UK, September 1994.
[3] J. Beran, R. Sherman, M. S. Taqqu, and W. Willinger. Long-range dependence in
variable bit rate video traffic. IEEE Transactions on Communications, 43:1566-
1579, April 1995.
[4] J. Beran, R. Sherman, M. S. Taqqu, and W. Willinger. Variable-bit-rate video
traffic and long-range dependence. IEEE Trans, on Communication, 1992.
[5] V. Paxson and S. Floyd. Wide area traffic: The failure of Poisson modeling.
IEEE/ACM Transactions on Networking, 3:320-328, 1995.
[6] R. S. Pazhyannur and R. Agrawal. Feedback based flow control in ATM networks
with mutliple propagation delays. In Proceedings of the Conference on Computer
Communications (IEEE Infocom), San Fransisco, California, March 1996.
[7] R. Pazhyannur and R. Agrawal. Analytical and numerical results for feedback
based flow control of BISDN/ATM networks with significant propagation de
lays. In Proceedings of the Conference on Computer Communications (IEEE
Infocom), Boston, Massachusetts, April 1995.
[8] D. Hunt. Credit-based FCVC proposal for ATM traffic management -revision
Rl. Technical Report 94-0168R1, ATM Forum, May 1994.
[9] A. Dailianas and A. Bovopoulos. Real-time admission control algorithms with
delay and loss guarantees in ATM networks. In Proceedings of the Conference
on Computer Communications (IEEE Infocom), Boston, Massachusetts, April
1995.
[10] K. S. Kim and B. G. Lee. Three-level traffic shaper and its application to source
clock frequency recovery for VBR video services in ATM networks. Transactions
on Networking, 3(4), August 1995.
[11] S. Y. Lu. Integrated Traffic Control Mechanisims for ATM Networks. PhD
thesis, University of Florida, 1994.
[12] S. Y. Lu and H. A. Latchman. Analysis and optimization of a partial buffer
sharing scheme for ATM switch overload control. Proceedings of the Multimedia
Computing Conference 1995, San Jose, pages 459-483, 1995.
116


16
Again, misinserted cells appearing in severely faulty blocks should be excluded. Cell
misinsertion on a particular connection is most likely caused by an undetected error
in the header of a cell being transmitted on a different connection. This perfor
mance parameter is defined as a rate since the mechanism producing mis-inserted
cells is independent of the number of transmitted cells received on the corresponding
connection.
2.2.2 Speed
The QoS classes regarding speed include Cell Transfer Delay (CTD) and Cell
Delay Variation (CDV). The CTD is defined as the elapsed time between a cell exit
from the source and the corresponding cell at destination. The CDV is also called
jitter. There are two performance parameters associated with CDV: 1-point CDV
and 2-point CDV. The 1-point CDV describes variability in the pattern of cell arrival
events observed at a single measurement point with reference to a negotiated peak
rate. The 2-point CDV describes variability in the pattern of cell arrival events
observed at the output of a connection portion with reference to the pattern of the
corresponding events observed at the input to the connection portion.
2.3 Traffic Characteristics
Network traffic can be characterized in many ways depending on the adapted
criterion. One of the most widely used criterion is the tolerance of information delay
or loss. For instance, real-time traffic, such as that for CBR service, is distinguished
by its stringent CTD and CDV requirements, whereas best effort traffic, such as
data transmission, demands superior CLR but yields inferior performance in terms
of CTD and CDV. Consequently, an improper call admission control algorithm could
not only underutilize network resources, but also make difficult the delivery of the
QoS of eligible users. Another example for traffic characteristics is that some CBR
traffic such as voice can tolerate a certain portion of cell discarding in a congested


75
Motion Pictures Expert Group defined by ISO. Compared to H.261, the MPEG
standard was designed for a higher range of rates and much better visual quality.
Although MPEG video is suitable for a large number of multimedia applications
including videoconferencing, distant learning, and video mail, it is not, however,
intended to be broadcast television quality which signals at 10 45 Mbps.
An MPEG coder generates three types of frames: Intraframes (I) that use
image compression scheme, Predicted (P), and Bidirectional (B) frames that use
compression with motion compensation. While P frames are coded based on the
previous frame, B frames are based on both a past and a future reference frame.
Therefore, I frames take advantage of spatial locality, and P and B frames exploit
temporal locality. The MPEG algorithm uses several parameters that, when varied,
can lead to a significant change in the characteristics of a video source. Two of these
parameters, lq, the quantization level and fi, the interframe to intraframe ratio, are
of particular interest as they can be used to control the video source via network
feedback [60]. For example, if the network detects a trend of congestion, a feedback
message could be sent to all video sources that would require them to decrease their
sending rate. This can be accomplished by decreasing lq at the expense of a temporary
loss of quality. Similarly, if the cell loss rate of the network temporarily increases,
then, by decreasing fi the frequency of intraframes in the MPEG coding sequence
could be increased. Since intraframes halt error propagation, this action will ensure
that the duration of error propagation is short.
Figure 6.1(a) shows the frame size over time elapsed of a typical MPEG com
pressed video with f\ equal to 29 (GOP=30), obtained from a recorded sequence of
TV advertisements. Figure 6.1(b) depicts the relative frequency of the frame size.
The encoding pattern is as follows.
IBBBBBBBBBPBBBBBBBBBPBBBBBBBBBIBB...


86
2.5
2
fl1-5
o
1
0.5
0 2 4 6 8 10
time (sec)
x 10
(a)
(b)
Figure 6.8: Cumulative number of cells sent and lost with no feedback control


4
In Chapter 5, we propose a preventive flow control scheme using a Connection
Admission Control (CAC) suited for generic self-similar traffic, particularly multime
dia traffic in ATM networks. This CAC takes into account the Hurst parameter of
incoming traffic sources. To facilitate the CAC, we derive the cell loss probability for
a queueing system with an fGn arrival process and deterministic service time. The
CAC is then implemented based on an upper bound of this cell loss probability. In
addition, a number of numerical results are provided to validate the effectiveness of
the proposed CAC scheme.
In Chapter 6, we propose an adaptive rate-based flow control scheme for real
time VBR traffic in ATM networks. The goal of the scheme is to minimize the
impact of traffic overload in order to limit the cell loss rate to an acceptable range
and also increase the network utilization. The proposed flow control scheme is based
on predicting the evolution of buffer occupancy over time using a Proportional-plus-
Integral-plus-Derivative (PID) controller and a linear predictor to adaptively update
the optimum data emission rate at the transmitter. The adaptive policy attempts to
keep the buffer occupancy for each virtual channel at a steady level and the simulation
results show that the proposed scheme works effectively against network congestion.
Along with the design of the new flow control scheme, we also develop a hierarchically
structured testbed to measure network performance and explore various flow control
schemes in ATM networks with diverse classes of incoming traffic.
In Chapter 7, we present three protocols which are playing key roles in the
success of the future Internet Multimedia Network: Resource Reservation Proto
col (RSVP), Real-time Transport Protocol/Real-time Transport Control Protocol
(RTP/RTCP), and IP Multicasting Protocol-MBone. To study network dynamics
and real traffic characteristics, we conducted traffic measurements and analysis based
on empirical traces collected in the Laboratory for the Information Systems and
Telecommunications (LIST) in August 1997. We illustrate traces standing for more


108
reach the dual goals of keeping cell loss rate low and network utilization high, we
proposed an adaptive rate-based flow control scheme for real-time VBR traffic in
ATM networks. In other words, the goal of the scheme was to minimize the impact
of traffic overload in order to limit the cell loss rate to an acceptable range and also
increase the network utilization.
The proposed flow control scheme was based on predicting the evolution of
buffer occupancy over time using a Proportional-plus-Integral-plus-Derivative (PID)
controller and a linear predictor to adaptively update the optimum data emission
rate at the transmitter. The adaptive policy attempts to keep the buffer occupancy
for each virtual channel at a steady level and the simulation results showed that the
proposed scheme worked effectively against network congestion. In addition to the
design of the new flow control scheme, we also developed a hierarchically structured
testbed to measure network performance and explore various flow control schemes in
ATM networks with diverse classes of incoming traffic.
8.4 Future Work
Future directions regarding this work are summarized as follows:
Traffic characterization:
The traces collected in the LIST represent typical network behaviors in a small
workgroup environment. The study of traffic characteristics in a larger environ
ment such as Department or Campus will be beneficial to the design of ATM
switches and flow control schemes in the backbone of wide area networks.
Preventive flow control scheme:
In the proposed CAC, each source has to submit the traffic parameters including
H, a2, H, and desired Cell Loss Ratio e. This CAC is, therefore, better suited
for applications such as Video on Demand (VOD). On the other hand, for the
CAC to be able to support real-time traffic, these parameters must be computed


88
(b)
Figure 6.10: Cumulative number of cells sent and lost with feedback control


89
N vbr
CLR
Pn
l
0.072
0.433
2
0.082
0.521
3
0.102
0.604
4
0.167
0.892
5
0.304
0.954
Table 6.2: CLR and normalized utilization rate vs. number of VBR VCs
streams get aggregated on the trunk link. Consequently, both CLR and p^ increase
as predicted.
6.6 Summary
In this chapter, we presented a closed-loop rate-based flow control scheme for
real-time VBR traffic in ATM networks using a methodology based on the classical
PID controller. The primary goal of the flow control scheme is to limit the CLR to an
acceptable range and substantially increase the network utilization rate by adjusting
the sending rate of the video stream at the transmitter dynamically based on the
network condition obtained from the feedback. The simulation results show that the
proposed scheme works quite well.


CHAPTER 5
PREVENTIVE FLOW CONTROL SCHEME FOR SELF-SIMILAR TRAFFIC
Recent measurement studies have shown that the burstiness of the traffic in
LANs or WANs is associated with long-range dependence (LRD). In Chapter 3, we
showed that LRD can be efficiently characterized by a queueing model with self
similar arrival processes such as fractional Gaussian noise (fGn), or autoregressive
integrated moving average (ARIMA) processes. As indicated in [1], the fGn provides
the simplest way to model the LRD and self-similarity. In this chapter, we will,
consequently, consider fGn as the arrival process in an ATM network environment.
To avoid the congestion and buffer overflow resulted from bursty traffic, we
propose a preventive flow control scheme which consists of a new call admission
control (CAC) algorithm. This CAC algorithm uses the Hurst parameter as an
index of burstiness and to characterize the self-similar nature of aggregate traffic in
virtual paths of ATM networks. To facilitate the efficiency of CAC for self-similar
traffic, we derive the cell loss probability for a queueing system with finite buffer and
fGn arrival process. Note that the derivation is served as an extended version of the
work proposed by I. Norros [54], The numerical results reveal that the proposed CAC
algorithm outperforms the peak rate allocation scheme by rendering higher statistical
multiplexing gains (SMG). In addition, the algorithm provides less deviation from the
target cell loss probability than other statistical allocation scheme such as Equivalent
Capacity.
58


28
drawback the incapacity of capturing the autocorrelation function of {In}, which
to a large extent explains the phenomenon of traffic burstiness.
3.1.2 Markov Process Models
Unlike Poisson process models, Markov process models introduce dependence
into the random sequence {In} so that they can potentially capture traffic burstiness.
Consider a continuous-time Markov process X(t) with a discrete state space i £
{0,1, 2,..., m}. Consequently, a set of random variables X : {Xi} form a Markov
chain, where the dependency extends backwards only one unit in time. In other
words, the way in which the entire past history affects the future of the process is
completely summarized in the current state of the process. This property is so called
Markov or memoryless property.
Figure 3.1 illustrates an example of 2-state Markov chain where X stays in a
state 0 for an exponentially distributed holding time (required to satisfy the Markov
property) with parameter Ao, and then jumps to state 1 with probability a. In steady
state, we have the state equation
Po
Pi
= Q
1 a /3
a 1 ¡3
Po
Pi
Po
Pi
(3.3)
where po and p\ are the steady-state probability of Xq and X\, respectively, and
Q is called a transition probability matrix. Note that in a simple Markov traffic
model, each jump of the Markov process X is interpreted as signaling an arrival, so
interarrival times are exponentially distributed with their rate parameters dependent
on the state from which the jump occurred. This situation results in dependence
among interarrival times as a consequence of the Markov property.


This dissertation was submitted to the Graduate Faculty of the College of
Engineering and to the Graduate School and was accepted as partial fulfillment of
the requirements for the degree of Doctor of Philosophy.
May 1998
Winfred M. Phillips
Dean, College of Engineering
Karen A. Holbrook
Dean, Graduate School


11
value Oil in the PT field suggests a user cell from AAL5 service class and having
experienced congestion, and the value 110 simply indicates a Resource Management
(RM) cell.
The CLP field may be used for loss priority indication by the ATM endpoint
and for selective cell discarding in network equipment. In a given ATM connection,
and for each user-data cell in the connection, the ATM equipment that first emits the
cell can set the CLP bit equal to 0 or 1. The CLP bit is used to distinguish between
cells of an ATM connection: A CLP bit equal to 0 indicates a higher priority cell and
a CLP bit equal to 1 indicates a lower priority cell which is subject to discard upon
network congestion.
The UNI and Network/Network interface (NNI) are similar, the difference
being that the UNI will connect access ports of ATM switches and Customer Premise
Equipment (CPE) which could include broadband terminals, terminal adaptors, and
cell-based LAN/WAN equipment, while the NNI can only connect trunk ports of
ATM switches. The NNI is intended for ATM subnetworks or networks and, hence,
does not need the GFC field in the ATM cell header. Moreover, the GFC field
has been subsumed into the VPI field, allowing 16 times as many virtual paths.
There exist a number of different UNI/NNI specifications, each of which is basically
a stack comprising one particular physical layer and the ATM layer. In addition,
the UNI/NNI encompasses signaling and connection management aspects from the
control plane.
2.1.4 ATM Adaptation Layer
From previous sections, the ATM layer not only uses the service of the phys
ical layer to transport cells, but also delivers cell payloads to the upper layers. In
most cases, it is necessary to perform some adaptation functions, for instance, to


44
3.4.2 Aggregation of Renewal Processes
A self-similar trace generator described in [39] consists of aggregating a num
ber of renewal rewards processes to obtain a self-similar process. This approach has
the problem that one must trade off speed for the degree of self-similarity, as in
creasing the Hurst parameter requires that the number of renewal rewards processes
be increased. Conversely, one advantage of this method is that it can be efficiently
implemented on a parallel computer. Leland and et al [39] report taking 3-5 minutes
to generate a trace of 100,000 points on a massively parallel computer with 16,384
processors using this method.
3.4.3 M/G/oo with Heavy-tailed Distributed Service Time
Consider an M/G/oo queue, where customers arrive according to a Poisson
process and have service times drawn from a heavy-tailed distribution. In this model,
the value of the sample path at time t is the number of customers in the system at
time t. This model produces sample paths which are asymptotically self-similar.
The drawback of this approach is that, once again, one must trade off speed for the
degree of self-similarity, but unlike the previous approach, this method is not easily
parallelized.
3.4.4 AR.IMA
This method described in [2] generates sample paths from a fractional ARIMA
process which are asymptotically self-similar. This algorithm has the major drawback
that it requires 0(n2) running time to generate a sample path of length n. The
authors of [2] report taking 10 CPU hours to generate 171,000 points using this
method.


114
7ohold on
axis([0 r -1 0]);
title(Variance-Time Plot)
ylabel(loglO (Normalized Variance))
xlabeKloglO (Aggregation Level (m) ))
% end of VTP Algorithm
B.2 Rescaled Adjustment Algorithm
7. Hurst parameter Estimator by RS
7. x: self-simialr sequence
7, h: target husrt parameter
function [rs, n]=rs(x,h)
na=max(size(x));
s=zeros(1000,1);
ni=inc;
j=i;
s(1)=1;
while s(j) j=j+i;
s(j)=l+(j-l)*ni;
end
count=j-l;
rs=zeros(count, 1);
w=zeros(count, 1) ;
for i=l:count,
for k=s(i):na,
w(k)=(sum(x(l:k)))-k*mean(x(l:s(i)));
end


85
(a)
(b)
Figure 6.7: Transmitted and received frame size with no feedback control
(Kj,Kd)
CLR
Pn
(0.00, 0.00)
0.046
0.301
(0.20, 0.00)
0.073
0.427
(0.20, 0.05)
0.072
0.433
(0.20, 0.15)
0.081
0.456
(0.00, 0.15)
0.053
0.358
(0.10, 0.15)
0.062
0.405
(0.15, 0.15)
0.068
0.418
Table 6.1: CLR and normalized utilization rate under various (K/, K0)


39
G(n) = BH(n + 1) BH(n) (3.28)
A fractional Gaussian noise (fGn) is a stationary Gaussian process with mean ¡jl.
variance cr2, and autocorrelation ftmction
r(r) (l/2)(|r + 1\2H 2\t\2H + \t- 1\2H) (3.29)
Prom Proposition 1, we prove that r(r) -> H(2H 1)\t\2H~2 as t -> oo, which
means fGn exhibits LRD It can also be shown that the aggregated processes X^
all have the same distribution as X for all 0 < H < 1. Thus, by (3.6), fGn is exactly
self-similar with Hurst parameter H and possesses LRD as long as 1/2 < Lf < 1.
Fractional Gaussian noise was originally introduced in 1968 by Mandelbrot
and Van Ness [31]. The applicability of fGn in traffic models is somewhat limited
by its strict autocorrelation structure. This limitation makes it less suitable for
modeling traffic streams which have very strong short-range dependence; for some
phenomenon, however, fGn can be a reasonable approximation. It should also be
noted that although fGn is a Gaussian process, an fGn sample path can be trans
formed into a sample path with arbitrary marginal distribution while maintaining
the Hurst parameter H [36].
3.3.3 Fractional ARIMA(p.d.q) Processes
Autoregressive-moving average (ARMA) processes are often used for modeling
empirical time series. Let p and q be non-negative integers, then the sequence {Xn}
is called ARMA(p,q) if it satisfies the equations
p Q
Xn ~ ^ ] Q/kXnk T ) ] bk^nki (3.30)
fc=1 fc=0
where {wk : k = 0,1,2,...} are i.i.d random variables and are assumed to be either
Gaussian or non-Gaussian. The {a^} are referred to as AR parameters while {b^}
are called MA parameters. For convenience, denote


EFFICIENT FLOW AND CONGESTION CONTROL
FOR
SELF-SIMILAR TRAFFIC
IN
ATM NETWORKS
BY
WEN-YEN FU
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1998


CHAPTER 7
INTEGRATED SERVICES PROTOCOLS AND TRAFFIC ANALYSIS
The Internet has proven to be the best solution for global sharing of infor
mation. Through the Internet, one can access information from world wide web
(WWW) and exchange electronic mail, as well as share applications with others from
any continent on earth. The success of these end-to-end communications is essen
tially achieved by two transport layer protocols in the Internet protocol suite, namely,
Transport Control Protocol (TCP) and User Datagram Protocol (UDP), where TCP
provides a connection-oriented, reliable flow between two hosts and UDP provides a
connectionless, unreliable datagram service.
Currently one class of service in the Internet exists normally referred to as
best effort whose traffic is characterized by first come first serve scheduling at each
node in the network. Best effort service has worked extremely well for traffic without
time constraint. However, for real-time traffic, such as voice and video, TCP/UDP
has performed well only when networks are underloaded. In order to provide guar
anteed quality of service (QoS) for real-time traffic ceaselessly, new classes of service
and new protocols are being introduced on the Internet while retaining the existing
best effort service. Three emerging technologies that will be able to support the
needed Internet multimedia services are Resource Reservation Protocol (RSVP),
Real-Time Transport Protocol (RTP), and IP Multicast Protocol.
Network traces have been widely considered as one of the most efficient ways
to study network dynamics, usage characteristics, and growth patterns. They are of
extreme importance to trace-driven simulations for tools like OPNET. In this Chap
ter, we describe the three real-time oriented protocols mentioned above in Session 1.
90


5
than 500,000 packets in different time scales. The traces are also treated separately
according to the protocols of interest to us, which are IP Multicasting, Transport
Control Protocol (TCP) and User Datagram Protocol (UDP). And finally, we con
clude the dissertation with summary and future directions in Chapter 8.


107
words, the flow control schemes which use Poisson-based traffic models resulted in
an overestimation of their performance.
8.2 Preventive Flow Control
Preventive flow control generally involves the following two procedures: call
admission control (CAC) and bandwidth enforcement. Since ATM is a connection-
oriented, a call setup procedure has to be carried out before a user starts transmitting
over an ATM networks. The main objective of this procedure was to establish a path
between the sender and the receiver. In addition, it also allocated resources in every
switch along the path to accepted connections. In short, CAC dealt with the question
whether or not a switch can accept a new connection.
For generic self-similar traffic, particularly multimedia traffic in ATM net
works, we proposed a preventive flow control scheme using a Connection Admission
Control (CAC). This CAC takes into account the Hurst parameter of incoming traffic
sources. To facilitate the CAC, we derived the cell loss probability for a queueing
system with an fGn arrival process. The CAC was then implemented based on an
upper bound of this cell loss probability. The numerical results showed that with the
proposed CAC, more connections can be accepted into the networks than with peak
rate allocation CAC. The simulations also indicate that the QoS is ensured with the
proposed CAC.
8.3 Feedback Flow Control
In communication networks with large delay-bandwidth product, congestion
could happen over shorter time scales than those at which end-to-end protocols such
as congestion control schemes typically operate. In such cases, the congestion can
dissipate rapidly before congestion feedback information returns to the source. Net
work designers, therefore, face a challenge. The bursty and cyclic nature of Variable
Bit Rate (VBR) traffic creates another issue for transmission in ATM networks. To