Title: Wavelength allocation strategies in optically switched networks for avionics
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00094722/00001
 Material Information
Title: Wavelength allocation strategies in optically switched networks for avionics
Physical Description: Book
Language: English
Creator: Reardon, Casey B.
Profumo, John D.
George, Alan D.
Publisher: Reardon et al.
Place of Publication: Gainesville, Fla.
Copyright Date: 2006
 Record Information
Bibliographic ID: UF00094722
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.


This item has the following downloads:

AVFOP2006 ( PDF )

Full Text

Casey B. Reardon, John D. Profumo, Alan D. George
High-performance Computing and Simulation (HCS) Research Laboratory
Department of Electrical and Computer Engineering, University of Florida
Gainesville, FL 32611-6200

1. Introduction
For advantages in performance, scalability,
protocol transparency, cost-effectiveness, etc.,
WDM-based optical networks are being evaluated
for use in emerging avionics systems. One area
where optical technologies are rapidly developing
is optical switching. Optical switches offer the
potential performance, scalability, and flexibility
that advanced avionics networks will demand in
the future. This paper investigates via simulation
modeling an optical switching architecture for
networking on advanced avionics platforms. Two
strategies for wavelength allocation are presented,
and their performance characteristics are
compared. Additional simulation experiments
analyze the effects of varying two architectural
parameters within each allocation strategy.

2. Background Information
The optical switching architecture evaluated in
this paper is based on the OSMOSIS architecture
developed by IBM for high-performance
computing systems [1]. Each point-to-point
connection includes an optical data path and a
separate electronic control path used to request
and reserve optical connections. Transmission
requests are made to the switch through the
control path. The switch arbiter responds to the
request indicating when the optical path is
available, and reserves the path in the optical
backplane. Data is then transmitted over the
optical connection for the number of time slots
allotted by the switch arbiter. At the end of each
time slot, the optical outputs are reconfigured, if
needed, for data transmission over the next time
slot. A broadcast-and-select design is used for the
optical switching. Each optical input is split and
distributed to all outputs. Each output then
chooses the desired optical input. The choice of
optical devices in this stage will heavily impact
switching speeds.

An OSMOSIS-based switching architecture
was modeled in the Library for Integrated Optical
Networks (LION). Developed at the University of
Florida, LION was created within a discrete-event
simulation environment called MLDesigner from
MLDesign Technologies. LION provides users
with the flexibility to design systems with a broad
range of optical devices and varying high-level
components. New and legacy network protocols
can be layered on top of optical components to
realize and evaluate almost any system design.

3. Case Study based on OSMOSIS
In this case study, two wavelength allocation
approaches are considered. The first approach,
fixed-destination assignment, allocates each
destination node a single wavelength. Optical
transmitters are thus responsible for tuning to the
correct wavelength of the destination for each
message. The effects of tuning delays are
minimized in this approach by performing the
tuning while the switch arbiter performs the
scheduling. The second approach, fixed-
transmitter assignment, allocates a fixed
wavelength to each transmitter. Detectors are
now responsible for tuning to the correct
wavelength. While tuning delays cannot be
overlapped with scheduling as before, efficient
multicasting is possible, as multiple outputs can
select a single input during any time slot.

Figure 1. Proposed LAN Topology

1 of2

An experimental setup designed to represent a
future avionics platform is used in our simulative
experiments. This setup is comprised of 97 nodes,
which generate a combination of bursty, random,
and continuous traffic, totaling 200 MB/s.
Message sizes are uniformly distributed between
1,000 and 30,000 bits. The traffic pattern of this
setup mimics a centralized avionics architecture,
where most information passes through a central
processing system. Results are obtained from one
second of traffic simulated in this system.
The layout of the network implementation is
illustrated in Figure 1. The various end-nodes of
this network are connected to one of eight
perimeter switches, which are interconnected by
three backplane switches. Each perimeter switch
is designed to accommodate up to 32 end-nodes.

4. Results
The mean packet latencies measured in our
simulative experiments are presented in Tables 1
and 2. For each approach, four different timeslot
periods are considered. Also, three values for the
maximum number of consecutive slots allotted to
a single node at once are used. In each case, 100
nanoseconds of every time slot is used for optical
switching. Laser transmitters operate at 2.5 Gb/s.

Table 1. Fixed-Destination Mean Latency (ps)
Maximum Timeslot Period (ns)
Slot Allotment 300 500 1,000 2,000
7 29.8 23.2 21.3 22.8
10 29.5 22.0 21.4 22.4
15 29.2 22.9 21.0 22.8

Table 2. Fixed-Transmitter Mean Latency (ps)
Maximum Timeslot Period (ns)
Slot Allotment 300 500 1,000 2,000
7 31.0 24.0 22.1 23.5
10 30.8 23.3 22.2 23.5
15 30.6 23.6 21.7 23.2

The results in Table 1 and Table 2 show that
the fixed-destination wavelength protocol offers
slightly better performance than the fixed-
transmitter protocol in these experiments. The
average difference is approximately 1 us, which is
also the optical tuning delay used in our models.
The primary reason for the difference is the ability
to overlap scheduling and optical device tuning.

Networks with broadcast and multicast traffic
would see increased benefits from the fixed-
transmitter strategy, where multiple nodes can
simultaneously receive the same transmission.
Within each wavelength assignment protocol,
the most obvious trend observed is the reduction
of packet latency as the timeslot period increases.
Since the time reserved in each slot for optical
switching is constant (100ns), a smaller fraction of
overall time is used performing optical switching
with longer timeslots. When the timeslot period is
increased to 2,000 ns, performance begins to
decrease. One factor is the underutilization of
allotted resources with small messages, which
only occupy a fraction of one timeslot.
Additionally, a minimum of two timeslots are
needed for scheduling any transaction. Thus, a
longer timeslot period increases the minimum
scheduling delay every packet observes.
Small gains in network performance were also
observed by increasing the maximum slot
allotment parameter. While a higher consecutive
timeslot allotment decreases the average message
latency, the theoretical maximum queueing delay
in the round-robin scheduler increases.

5. Conclusions
In this paper, an architecture for an optically-
switched avionics network is presented. The
performance of this architecture was analyzed
using simulative experiments for two wavelength
allocation strategies. Additionally, the effects of
varying the timeslot period and maximum number
of consecutively allotted timeslots were analyzed.
Results showed slightly lower packet latencies
with a fixed-destination wavelength protocol.
Larger timeslot periods, up to 1,000 ns, also
improved performance, although further increases
would not be beneficial. The optimal parameters
for any platform will always depend upon the
nature of the network traffic. Our modeling tools
and approach allow us to evaluate such design
decisions for a wide range of network scenarios.

6. References
[1] Hemenway, R., and R. Grzybowski, "Optical
Packet-Switched Interconnect for
Supercomputer Applications," Journal of
Optical Networking, Vol. 3, No. 1, Dec. 2004.

2 of 2

University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs