<%BANNER%>

Mobility Management for Wireless Mobile Networks


PAGE 1

MOBILITY MANA GEMENT F OR WIRELESS MOBILE NETW ORKS By WENCHA O MA A DISSER T A TION PRESENTED TO THE GRADUA TE SCHOOL OF THE UNIVERSITY OF FLORID A IN P AR TIAL FULFILLMENT OF THE REQUIREMENTS F OR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORID A 2003

PAGE 2

Cop yrigh t 2003 b y W enc hao Ma

PAGE 3

T o m y wife, Xinglin Liu, for her supp ort, encouragemen t and lo v e

PAGE 4

A CKNO WLEDGMENTS I w ould lik e to express m y gratitude to all those who ha v e help ed me to complete this dissertation. I am deeply indebted to m y advisor, Prof. Y uguang F ang, for his moral supp ort and encouragemen t through the y ears for Ph.D study I also w an t to thank m y committee mem b ers, Prof. Alan D. George, Prof. Chien-Liang Liu, Prof. Janise McNair and Prof. Sarta j. K. Sahni, for their in terests and commen ts. Sp ecial thanks go to m y colleagues in the Wireless Net w orks Lab oratory (WINET) for their help and supp ort in m y researc h w ork. This material is based up on w ork supp orted b y the National Science F oundation under Gran t No. ANI-0093241. An y opinions, ndings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily rerect the views of the National Science F oundation. iv

PAGE 5

T ABLE OF CONTENTS page A CKNO WLEDGMENTS . . . . . . . . . . . . . . . . iv LIST OF T ABLES . . . . . . . . . . . . . . . . . . vii LIST OF FIGURES . . . . . . . . . . . . . . . . . . viii ABSTRA CT . . . . . . . . . . . . . . . . . . . . x CHAPTERS 1 INTR ODUCTION . . . . . . . . . . . . . . . 1 1.1 Mobilit y Managemen t for Wireless Mobile Net w orks . . . . 1 1.2 Mobilit y Managemen t Sc hemes in Curren t Systems . . . . . 2 1.3 Curren t Mobilit y Managemen t Researc h . . . . . . . . 4 1.4 Organization of the Dissertation . . . . . . . . . . 6 2 POINTER F OR W ARDING BASED MOBILITY MANA GEMENT STRA TEGIES . . . . . . . . . . . . . . . . . . 7 2.1 In tro duction . . . . . . . . . . . . . . . 7 2.2 Signaling Net w ork Arc hitecture . . . . . . . . . . 9 2.3 Basic Mobilit y Managemen t Pro cedures of IS-41/GSM MAP . . 10 2.4 Tw o-Lev el P oin ter F orw arding Sc heme . . . . . . . . 11 2.4.1 Mobilit y Managemen t Pro cedures . . . . . . . 11 2.4.2 Cost F unctions and P erformance Analysis . . . . . 14 2.4.3 P erformance Analysis under Exp onen tial RA Residence Time 19 2.4.4 Sensitivit y Analysis to the Residence Time . . . . . 22 2.5 POFLA Sc heme . . . . . . . . . . . . . . 23 2.5.1 The Sc heme Ov erview . . . . . . . . . . . 23 2.5.2 System Mo del and Cost F unctions . . . . . . . 25 2.5.3 P erformance Ev aluations . . . . . . . . . . 31 2.5.4 Sensitivit y to the RA Residence Time . . . . . . 36 2.6 Discussions . . . . . . . . . . . . . . . . 37 2.7 Conclusions . . . . . . . . . . . . . . . 38 3 USER PR OFILE BASED STRA TEGY . . . . . . . . . . 39 3.1 In tro duction . . . . . . . . . . . . . . . 39 3.2 System Description . . . . . . . . . . . . . 40 3.3 PBS and MPBS Sc hemes . . . . . . . . . . . . 41 3.3.1 Profle Based Lo cation Sc heme . . . . . . . . 41 3.3.2 Mobilit y P attern Based Sc heme . . . . . . . . 42 3.4 Cost Ev aluation and Sim ulations . . . . . . . . . . 47 3.5 Numerical Results and Comparison . . . . . . . . . 51 v

PAGE 6

3.6 Conclusions . . . . . . . . . . . . . . . 56 4 MOBILITY MANA GEMENT IN MOBILE IP NETW ORK . . . . 58 4.1 In tro duction . . . . . . . . . . . . . . . 58 4.2 Related W orks . . . . . . . . . . . . . . . 61 4.3 Dynamic Hierarc hical Lo cation Managemen t Sc heme . . . . 63 4.4 Analytical Mo del . . . . . . . . . . . . . . 65 4.5 Numerical Results . . . . . . . . . . . . . . 70 4.6 Comparison with the IETF Hierarc hical Sc heme . . . . . 77 4.7 Impro v emen t of DHMIP Sc heme . . . . . . . . . . 79 4.7.1 State Activ ation . . . . . . . . . . . . 79 4.7.2 Lo op Remo v al . . . . . . . . . . . . . 81 4.8 Conclusions . . . . . . . . . . . . . . . 82 5 CONCLUSIONS AND FUTURE W ORKS . . . . . . . . . 84 REFERENCES . . . . . . . . . . . . . . . . . . . 87 BIOGRAPHICAL SKETCH . . . . . . . . . . . . . . . . 92 vi

PAGE 7

LIST OF T ABLES T able page 4.1 The dynamic hierarc hical Mobile IP proto col . . . . . . . . . 65 4.2 The p erformance analysis parameters (1) . . . . . . . . . . 71 4.3 The p erformance analysis parameters (2) . . . . . . . . . . 71 vii

PAGE 8

LIST OF FIGURES Figure page 2.1 Reference CCS net w ork arc hitecture . . . . . . . . . . . 9 2.2 The TwoL evelFwdMO VE pro cedures . . . . . . . . . . . 11 2.3 The TwoL evelFwdFIND pro cedures . . . . . . . . . . . 12 2.4 Relativ e MO VE and FIND costs of forw arding with = 0 : 3 ; K = 1 : 5 . . 20 2.5 Relativ e MO VE and FIND costs of forw arding with = 0 : 3 ; K = 4 . . . 21 2.6 Relativ e MO VE and FIND costs of forw arding with = 0 : 6 ; K = 1 : 5 . . 22 2.7 The eect of v ariance in residence time ( r ) with = 0 : 3 ; K = 1 : 5 ; K 1 = 4 and K 2 = 4 . . . . . . . . . . . . . . . . . 23 2.8 POFLA strategy pro cedures . . . . . . . . . . . . . 24 2.9 The relativ e costs for the three sc hemes with P = 0 : 05, G = 0 : 1 and f = 1 : 5 32 2.10 The relativ e costs for the three sc hemes with P = 0 : 05, G = 0 : 1 and f = 3 33 2.11 The relativ e costs for the three sc hemes with P = 0 : 1, G = 0 : 1 and f = 1 : 5 34 2.12 The relativ e costs for the POFLA and DLA with P = 0 : 05, G = 0 : 1 and f = 1 : 5 . . . . . . . . . . . . . . . . . . 34 2.13 The relativ e net costs for the POFLA and SLA with P = 0 : 05, G = 0 : 1 and f = 1 : 5 . . . . . . . . . . . . . . . . . . 35 2.14 The eect of v ariance of residence time ( r ) with P = 0 : 05, G = 0 : 1 and f = 1 : 5 36 3.1 Mobilit y pattern based sc heme pro cedures . . . . . . . . . 43 3.2 The up date n um b ers for three sc hemes in 24 hours with user residence time 30 min utes . . . . . . . . . . . . . . . . . 46 3.3 Sim ulation net w ork arc hitecture . . . . . . . . . . . . 49 3.4 The lo cation up date cost ratio of MPBS sc heme to IS-41/GSM MAP . . 50 3.5 The comparison of the lo cating time for MPBS and PBS . . . . . 51 3.6 The total costs of MPBS and PBS to IS-41/GSM MAP with uniform distribution 52 3.7 The total costs of MPBS and PBS to IS-41/GSM MAP with linear distribution 53 3.8 The total costs of MPBS and PBS to IS-41/GSM MAP with exp onen tial distribution . . . . . . . . . . . . . . . . . 53 viii

PAGE 9

3.9 The eects of user out-of-pattern probabilit y on MPBS and PBS with pag ing cost = 0 : 1 ; C M R = 1 . . . . . . . . . . . . . . . 55 3.10 The eects of user out-of-pattern probabilit y on MPBS and PBS with pag ing cost = 0 : 5 ; C M R = 1 . . . . . . . . . . . . . . . 55 4.1 The MIP lo cation registration and pac k et routing . . . . . . . 60 4.2 The DHMIP lo cation registration and pac k et deliv ery . . . . . . 64 4.3 Comparison of the total costs for dieren t sc hemes . . . . . . . 72 4.4 Comparison of the total costs for uniformly distributed under dieren t parameters . . . . . . . . . . . . . . . . . 73 4.5 Comparison of the total costs under dieren t distributions . . . . 74 4.6 Comparison of the total costs under dieren t K . . . . . . . . 74 4.7 Comparison of the total costs under dieren t distributions with optimal K 75 4.8 The optimal n um b er of hierarc h y lev els for uniform distribution . . . 75 4.9 The eect of the v ariance of the F A residence time ( r ) . . . . . . 76 4.10 The comparison with the IETF hierarc hical sc heme . . . . . . . 79 4.11 The total cost for the enhanced DHMIP sc heme . . . . . . . . 81 4.12 The total cost with lo op r emoval . . . . . . . . . . . . 83 ix

PAGE 10

Abstract of Dissertation Presen ted to the Graduate Sc ho ol of the Univ ersit y of Florida in P artial F ulllmen t of the Requiremen ts for the Degree of Do ctor of Philosoph y MOBILITY MANA GEMENT F OR WIRELESS MOBILE NETW ORKS By W enc hao Ma August 2003 Chair: Y uguang \Mic hael" F ang Ma jor Departmen t: Electrical and Computer Engineering Mobilit y is an imp ortan t c haracteristic of wireless mobile net w orks. The lo cation of a called mobile user m ust b e determined within a certain time limit b efore an y service can b e deliv ered. Th us, mobilit y managemen t is a k ey comp onen t for the wireless net w orks to eectiv ely deliv er wireless in ternet services. In the curren t wireless systems, the mobilit y managemen t is accomplished b y a t w o-tier hierarc hical arc hitecture consisting of Home Location Register (HLR) and Visitor Lo cation Register (VLR). With the increase of the user p opulation and service t yp es, the curren t mobilit y managemen t sc heme faces serious c hallenges. Man y researc h w orks ha v e b een carried out to impro v e the mobilit y managemen t pro cedures for the future generation wireless comm unication net w orks. In our researc h, w e prop ose a few new mobilit y managemen t sc hemes based on either p oin ter forw arding or user mobilit y patterns. In the p oin ter forw arding based sc hemes, the user mobilit y managemen t op eration is impro v ed b y t w o kinds of p oin ter setup pro cedures and an additional lev el of managemen t en tit y|Mobilit y Agen t (MA). The MAs can b e distributed among the net w ork based on the curren t trac load and user mobilit y so that the new sc hemes are more adaptiv e and robust while minimizing the total signaling trac. Motiv ated b y the observ ation x

PAGE 11

that man y mobile users follo w some daily routines whic h can b e used to predict the users' lo cations, w e prop ose a user mobilit y pattern based sc heme called MPBS. In this sc heme, the users' lo cations can b e determined with less signaling exc hange according to the user proles. W e sho w that this sc heme can signican tly reduce signaling trac. Our third researc h thrust is the mobilit y managemen t for Mobile IP net w orks. Curren tly the mobilit y o v er the In ternet is addressed b y the Mobile IP proto col. In our researc h, w e presen t a dynamic hierarc hical Mobile IP mobilit y managemen t strategy (DHMIP). In the DHMIP sc heme, an F A hierarc h y is set up dynamically among the net w ork and is sp ecic for ev ery mobile user. W e also dev elop a rigorous analytical mo del to ev aluate the p erformance. The results sho w that the DHMIP sc heme solv es the hea vy lo cation up date trac problem in Mobile IP proto col eectiv ely xi

PAGE 12

CHAPTER 1 INTR ODUCTION 1.1 Mobilit y Managemen t for Wireless Mobile Net w orks The wireless mobile comm unication net w orks and the In ternet ha v e b een exp eriencing explosiv e gro wth since the early 1990s, and it is b eliev ed that the user p opulation will b e steadily increasing in the future [1{ 4]. The wireless net w ork is the only w a y to ac hiev e ubiquitous comm unication and computing [5]. The ample and ric h information on the In ternet fuels the demands for accessing in ternet applications through wireless net w orks. The small size and high pro cessing capabilit y of the user equipmen t mak e this p ossible. Unlik e the ordinary wired net w orks, there is not a xed relationship b et w een a mobile terminal ID and its lo cation in wireless net w orks. In the wireless comm unication net w orks, users can mo v e to an ywhere in the service area co v ered b y the net w orks. The terminal IDs do not implicitly pro vide the lo cation information of the users. In order to pro vide the users with services, there m ust b e some w a y for systems to lo cate the users ecien tly This is the concept of mobilit y managemen t in wireless comm unication systems [6]. Generally sp eaking, the mobilit y managemen t includes lo cation registration (up date) and call deliv ery [7, 8]. With the dramatic increase of the user p opulation, the signaling trac generated b y the curren t mobilit y managemen t strategies will consume a lot of system resources (wireless and wired bandwidth). F or the future generation wireless comm unication systems, the lo cation areas (LAs) b ecome smaller than b efore in order to increase the system capacit y [9]. Smaller LA size can help systems utilize radio sp ectrum more ecien tly and reduce the pac k et transmission latency more signican tly Ho w ev er, this also mak es the mobilit y managemen t signaling trac w orse [10]. The Mobile IP is the standard enabling terminal mobilit y o v er the In ternet [11]. The IP proto col has b een designed for wired net w orks. There are t w o ma jor functions for the terminal IP addresses in the In ternet. An IP address is used to iden tify a particular end system in the whole net w ork and is also used to nd a route b et w een the endp oin ts. The 1

PAGE 13

2 mobilit y-enabling proto col for the In ternet, Mobile IP extends the IP proto col b y allo wing t w o IP addresses asso ciated with one mobile host. A mobile host is assigned a p ermanen t address in its home net w ork and can obtain a temp orary one in the visited net w ork. In order to guaran tee con tin uing services, the mobile host is required to register with its home agen t to set up a binding b et w een the t w o IP addresses. Ho w ev er, Mobile IP is not a go o d solution for users with high mobilit y [12, 13]. The lo cation up date and registration messages can result in hea vy trac burden to the In ternet. Moreo v er, if a user is far a w a y from his/her home net w ork or the home agen t pro cessing capabilit y is o v erwhelmed b y the h uge v olume of up date messages, the signaling dela y for the lo cation up date could b e in tolerable, leading to the p ossible loss of a large amoun t of in-righ t pac k ets. In this dissertation, w e fo cus on mobilit y managemen t researc h for wireless mobile net w orks and attempt to solv e the aforemen tioned problems in more ecien t w a ys. In what follo ws, w e use the lo cation area (LA) and registration area (RA) in terc hangeably b ecause they are the t w o terms used in the IS-41 and GSM MAP proto cols for the same concept. 1.2 Mobilit y Managemen t Sc hemes in Curren t Systems In wireless cellular net w orks, a user's curren t lo cation m ust b e determined rst b efore an y service can b e deliv ered to the corresp onding wireless access p oin t. In suc h net w orks, the service area is divided in to cells. Eac h cell is primarily serv ed b y one base station, although a base station ma y serv e one or more cells [14]. Base stations are the users' radio p orts. An RA consists of an aggregation of a n um b er of cells, forming a con tiguous geographical region. The lo cation managemen t proto cols consist of t w o ma jor parts: lo cation registration (or lo cation up date) and call deliv ery In the lo cation registration pro cedure, the mobile terminal up dates its curren t lo cation information to some net w ork databases, and the information can b e retriev ed for the future call deliv ery pro cedure. Curren tly b oth the IS41 and GSM MAP share the same c haracteristics and b oth use a t w o-tier system consisting of Home L o c ation R e gister (HLR) and Visitor L o c ation R e gister (VLR) databases. The signaling net w ork used to set up calls is distinct from the net w ork used to actually transp ort the information con ten ts of the calls. It is the signaling net w ork, connected to the databases, that accomplishes the lo cation managemen t task for the wireless comm unication net w ork. In real systems, a VLR can b e in c harge of one or m ultiple RAs dep ending on the user densit y

PAGE 14

3 Without loss of generalit y w e assume that there is one VLR for ev ery RA in our study A VLR stores the proles of the users who are curren tly residing in its c harge area. The HLR stores the users' proles and the IDs of curren t serving VLRs in addition to some en tries suc h as securit y and billing information. According to the curren t mobilit y managemen t strategy a mobile terminal k eeps monitoring the broadcast signal con taining the ID information of cells and RAs If the mobile terminal nds that it c hanges RA, it will send a lo cation up date message to the new VLR via the curren t serving base station. Up on receiving the up date message, the new VLR will forw ard it to the user's HLR to request the user's prole and nish the AAA pro cedures at the same time. The HLR will up date the user's serving VLR ID in its database and send a de-registration message to the user's old VLR. The old VLR deletes the user's en try in its database and resp onds with an ac kno wledgmen t message to the HLR. In the call deliv ery pro cedure, when a call for this user is initiated b y some other caller, the lo cation request message is sen t to the user's HLR to nd out the user's curren t lo cation. The HLR forw ards the message to the user's curren t serving VLR. The curren t VLR can then determine the user's curren t serving cell b y paging all the cells in its c harging area. After nding out the user's curren t access p oin t (the BS), the VLR sends a resp onding message bac k to the HLR with a temp orary n um b er allo cated to the user for routing purp ose. Then, a trac c hannel can b e set up b et w een the caller and the callee after the HLR receiv es the n um b er and forw ards it to the caller's serving switc h. It is ob vious that, in this sc heme, the lo cation up date trac will increase dramatically with the increase of the user p opulation and the reduction of the RA size. The HLR ma y b ecome a b ottlenec k for the wireless net w orks [15]. The dierences of the net w ork organization b et w een wireless comm unication net w ork and the In ternet in tro duce dierences in the w a y the user mobilit y is dealt with. The user mobilit y o v er the In ternet is accomplished through the Mobile IP proto col [16]. In the Mobile IP proto col, a mobile host has a p ermanen t address (home address) registered in its home net w ork and this IP address remains unc hanged when the user mo v es from subnet to subnet. A home agen t is a router in a mobile host's home net w ork, whic h can in tercept and tunnel the pac k ets for the mobile host and also main tains the curren t lo cation information for the mobile host. If a mobile host roams to a foreign net w ork, the mobile host can obtain a new

PAGE 15

4 IP address from the foreign agen t. The new address is the mobile host's care-of address (CoA) used for pac k et routing purp ose. The CoA for the mobile host will c hange from subnet to subnet. In order to main tain con tin uous services while the user is on the mo v e, Mobile IP requires the mobile hosts to up date their lo cations to the home agen ts whenev er they mo v e to a dieren t subnet so that the home agen ts can in tercept the pac k ets and tunnel them to the user's curren t p oin ts of attac hmen t. Th us, the Mobile IP can ac hiev e con tin uous in ternet access services for mobile users and do es pro vide a simple and scalable solution to user mobilit y 1.3 Curren t Mobilit y Managemen t Researc h The curren t researc h activities in wireless comm unication net w orks generally fall in to t w o categories based on the system database arc hitecture [17, 18]. In the rst category the mobilit y managemen t strategies are dev elop ed, whic h aim at impro ving IS-41/GSM MAP sc heme while k eeping the basic database net w ork arc hitecture unc hanged. The adv an tage of this t yp e of solution is that it is easily adopted b y the curren t wireless comm unication net w orks without ma jor mo dication. These sc hemes are based on cen tralized database arc hitectures inherited from the IS-41/GSM MAP standard. The follo wing sc hemes b elong to this category: dynamic hier ar chic al datab ase ar chite ctur e [19]; p er-user lo c ation c aching [20]; user pr ole r eplic ation [21]; lo c al anchoring [22] and p ointer forwar ding [23]. Another category of researc h results lie in completely new database arc hitectures that require a new set of sc hemes for lo cation registration and call deliv ery The ma jor sc hemes b elonging to this category are: ful ly distribute d r e gistr ation scheme [24]; p artitioning [25] and datab ase hier ar chy [26]. All the ab o v e sc hemes can b e considered as static b ecause the mobile terminal is mandatorily required to register its curren t lo cation to some database ev ery time it en ters a new LA. Some sc hemes are prop osed for the mobile terminal to p erform lo cation registration dynamically [27{ 29]; in other w ords, the mobile terminal sends a lo cation update message to the system according to the user's curren t mo v emen t status [30]. Usually the dynamic sc hemes can ac hiev e b etter p erformance than static ones; ho w ev er they are more complicated in implemen tation to o. Sev eral imp ortan t dynamic mobilit y managemen t sc hemes are dynamic LA management [31]; time-b ase d, movement-b ase d and distanc e-b ase d dynamic up date schemes [32{ 34]. In our w ork, the prop osed two-level p ointer forwar ding

PAGE 16

5 and POFLA strategies b elong to the static sc heme category and the user mobility p attern b ase d scheme can b e considered as dynamic. In general, there is a tradeo b et w een the paging cost and the connection setup dela y [35, 36]. So, there are some researc h w orks in this area attempting to minimize the paging cost under a giv en dela y constrain ts, suc h as p aging under delay c onstr aints [37] and up date and p aging under delay c onstr aints [38]. The IP net w ork administration is based on the net w ork administrativ e regions. So the IP mobilit y can b e broadly classied in to macro-mobilit y and micro-mobilit y The macromobilit y is for the case that an MH roams across dieren t administrativ e domains. The macro-mobilit y o ccurs less frequen tly and usually in v olv es larger timescales [12]. The Mobile IP can ensure that the mobile users reestablish comm unication connections after a mo v e during macro-mobilit y The micro-mobilit y means the MH mo v emen t across m ultiple subnets within a single net w ork domain. F or micro-mobilit y whic h o ccurs quite often, the Mobile IP paradigm needs to b e enhanced. Most of the related w orks attempt to impro v e the Mobile IP micro-mobilit y handling capabilit y [13]. The micro-mobilit y prop osals can b e classied in to t w o distinct categories [39]. The rst is the r outing up date approac hes, in whic h IP routing is used to direct trac to w ard the MHs so that the m ultiple reencapsulation and decapsulation along the F As are a v oided. This category includes Hando-A war e Wir eless A c c ess Internet Infr astructur e (HA W AI I), T eleMIP and Cel lular IP The HA W AI I is a separate routing proto col to handle micro-mobilit y [40]. The sc heme hinges on the assumption that most user mobilit y is lo cal to an administrativ e domain of the net w orks. An MH en tering a new foreign net w ork is assigned a new CoA and retains its CoA unc hanged while mo ving within the foreign domain. F or macro-mobilit y the HA W AI I uses the traditional Mobile IP In this sense, the sc heme can b e considered as an enhanced Mobile IP The T eleMIP [12] sc heme tries to lo calize the scop e of lo cation up date messages b y in tro ducing a new managemen t en tit y In the T eleMIP a mobilit y agen t is in c harge of one region, handling the CoA addresses for those MHs roaming in the region. T eleMIP can enhance the p erformance in the IP supp ort o v er the cellular net w orks. The Cel lular IP scheme [41] is another strategy supp orting lo cal mobilit y in cellular en vironmen t. In this sc heme, the net w ork is connected to the In ternet through gatew a y routers and the roaming b et w een gatew a ys is managed b y Mobile IP while the mobilit y within access net w orks is handled b y a new designed proto col called cellular

PAGE 17

6 IP Another category of the micro-mobilit y sc hemes is the hier ar chic al tunnel approac hes, c haracterized b y their reliance on a tree-lik e structure of F As. The regional registration proto col b elongs to this category The Mobile IP r e gional r e gistr ation pr oto c ol [42] is an extension of the Mobile IP sc heme. The new proto col emplo ys the F A hierarc h y to lo calize the registration trac. In the regional registration sc heme, the net w ork arc hitecture is cen tralized. The distribute d dynamic r e gional lo c ation management scheme [43] is a dynamic hierarc h y strategy and the regional net w ork size is adjusted based on the user's curren t trac load and mobilit y information. This strategy can b e considered as an extension of the IETF regional registration sc heme to mak e it more rexible and adaptiv e. In our researc h, w e prop ose a new lo cation managemen t sc heme for Mobile IP net w ork| dynamic hier ar chic al Mobile IP (DHMIP). The new sc heme arc hitecture is dynamic and the F A hierarc h y is v arying based on the user b eha viors. 1.4 Organization of the Dissertation This dissertation is divided in to v e c hapters. F ollo wing this c hapter, c hapter 2 describ es the t w o-lev el p oin ter forw arding and POFLA strategies. In this c hapter, w e analyze the p erformance of the new sc hemes and compare them with the IS-41/GSM MAP sc heme theoretically W e also compare the p erformance of the t w o new sc hemes with those of the lo cal anc hor and p er-user forw arding sc hemes in this c hapter. A t the end of c hapter 2, w e briery discuss the relationship b et w een our new sc hemes and those previously men tioned and giv e some alternativ e w a ys for implemen tation. In c hapter 3, a new mobilit y managemen t sc heme, MPBS, for next generation wireless comm unication system is prop osed. In this sc heme, the mobile terminal will p erform lo cation up date based on the user mobilit y pattern. In this c hapter, w e describ e the details of the PBS and the new MPBS sc hemes and the p erformance ev aluations are carried out. Mobile IP is the proto col ac hieving mobilit y o v er the IP net w ork. Ho w ev er, the micro-mobilit y in Mobile IP proto col is not ecien t for most applications. Therefore w e in tro duce a new lo cation managemen t sc heme for Mobile IP net w ork in c hapter 4 and an analytical mo del is dev elop ed in that part to ev aluate the p erformance. In the last c hapter, w e presen t the conclusions and future researc h w ork in this area.

PAGE 18

CHAPTER 2 POINTER F OR W ARDING BASED MOBILITY MANA GEMENT STRA TEGIES 2.1 In tro duction In this c hapter, the t w o new lo cation managemen t strategies for wireless comm unication net w orks based on p oin ter forw arding tec hnology will b e presen ted. According to the IS-41/GSM MAP strategies, a mobile terminal p erforms lo cation update (registration) to the HLR ev ery time the user crosses a Registration Area (RA) b oundary and deregisters at the previous VLR [7, 8]. These lo cation managemen t pro cedures will incur high signaling trac. If man y users are far a w a y from their HLRs, hea vy signaling trac o v er the net w ork can o ccur. This problem b ecomes w orse with the increase of the n um b er of mobile users. The lo cal anc hor sc heme reduces the signaling trac b y c ho osing a lo cal anc hor for eac h user [22]. In this sc heme, a VLR close to a user is selected as the lo cal anc hor for him or her. Whenev er the user mo v es from one RA to another, the mobile terminal will p erform lo cation up date to the lo cal anc hor. The lo cal anc hor for a mobile user will not c hange unless a call request arriv es; at the same time, the HLR is also up dated via the call deliv ery pro cedures. When a call request terminating at this user is receiv ed b y the HLR, the user can b e traced b y the lo cal anc hor. The lo cal anc hor sc heme a v oids up date to HLR completely at the exp ense of increasing the lo cal signaling trac. Some similar sc heme w as prop osed in [44]. The dra wbac k of this kinds of sc hemes is that when the user k eeps mo ving constan tly without receiving an y call, the up dates to lo cal anc hor ma y b ecome costly to o, a similar b ottlenec k as the HLR is. F or example, at the end of conferences or games, man y p eople mo v e a w a y from one site without receiving calls, and the lo cal anc hor for these p eople can b ecome a b ottlenec k. Jain and Lin prop osed another sc heme called p er-user p oin ter forw arding sc heme [23]. In this sc heme, some up dates to the HLR can b e replaced b y setting up forw arding p oin ters from the previous VLRs to the new VLRs. When a call request to a mobile user arriv es, the wireless comm unication net w ork rst queries the user's HLR to determine the VLR whic h the user w as visiting at the previous lo cation up date, 7

PAGE 19

8 then follo ws a c hain of forw arding p oin ters to the user's curren t VLR to lo cate the mobile user. The trac to the HLR is decreased b y the p oin ter c hain; the p enalt y is the time dela y for tracing the user when a call to the user arriv es. The longer the p oin ter c hain, the less the signaling trac and the longer the call setup dela y T o mak e sure the setup dela y is under some constrain t, a threshold of the p oin ter c hain length is set. The user needs to p erform registration to the HLR whenev er the c hain threshold is reac hed. In order to o v ercome the dra wbac ks of the ab o v e t w o sc hemes, w e prop ose t w o kind of lo cation managemen t sc hemes based on the p oin ter forw arding tec hnique. One is two-level p ointer forwar ding str ate gy [45]; another is POFLA sc heme [46]. In b oth of the t w o sc hemes, t w o kinds of p oin ters are used. Some VLRs are selected as the Mobilit y Agen ts (MAs), whic h will b e resp onsible for the lo cation managemen t in a larger area comparing with the RAs and can b e geographically distributed. The MAs are also in v olv ed with longer or higher lev el p oin ters whic h can minimize the calling setup dela y The shorter or lo w lev el p oin ters c hains are set up b et w een the adjacen t VLRs. Both sc hemes can a v oid the p ossible costly up dates to HLR and the trac congestions in lo cal anc hor. More imp ortan tly the thresholds for the new sc hemes are t w o parameters whic h can pro vide the rexibilit y in design comparing with the one-parameter traditional p oin ter forw arding strategy The t w o sc hemes are detailed in the follo wing sections separately This rest of this c hapter is organized as follo ws. In section 2.2, w e describ e the basic signaling net w ork arc hitecture to facilitate the presen tation and analysis of the basic IS-41/GSM MAP proto cols, the new t w o-lev el p oin ter forw arding and POFLA strategies. Section 2.3 in tro duces the basic IS-41/GSM MAP lo cation managemen t strategy in detail. W e in tro duce and analyze the p erformance of the t w o-lev el p oin ter forw arding sc heme in section 2.4; w e also analyze ho w the user RA residence time can aect the t w o-lev el forw arding sc heme p erformance in this section. Another new p oin ter forw arding based lo cation managemen t sc heme, POFLA, is in tro duced and analyzed in section 2.5. In this section, w e also compare the the p erformances of the new sc hemes and some other prop osed strategies under v arious conditions. W e briery discusses the relationships b et w een our new sc hemes and some impro v ed ones in section 2.6, and section 2.7 giv es the conclusions.

PAGE 20

9 SSP/MSC VLR/ SCP/HLR Remote A link Local A link LSTP RSTP D linkRegistration Area Figure 2.1: Reference CCS net w ork arc hitecture 2.2 Signaling Net w ork Arc hitecture The net w ork used to accomplish signaling exc hange is distinct from the net w ork used to actually transp ort the information con ten ts of the calls. Esp ecially w e assume a Common Channel Signaling (CCS) net w ork is used to set up calls whic h use the Signaling System No.7 (SS7) proto cols [47]. Fig. 2.1 sho ws a t ypical No.7 signaling net w ork arc hitecture. In a CCS No.7 signaling net w ork, all the base stations in an RA are connected via wire-line net w ork to an end-oce switc h, or Service Switc hing P oin t (SSP). Eac h SSP serv es an RA. All the SSPs of dieren t RAs are in turn connected to a higher hierarc hical Lo cal Signaling T ransfer P oin ts (LSTP), whic h are connected to a Regional STP (RSTP). An RSTP connects to all the LSTPs in one region. In practice, eac h STP actually consists of t w o STPs in a matedpair conguration for reliabilit y F or the simplicit y of presen tation, Fig. 2.1 only sho ws one of eac h pair. The RSTPs are also connected to a Service Con trol P oin t (SCP). Eac h SCP is equipp ed with an HLR database. F or simplicit y w e assume that eac h VLR is asso ciated with one Mobile Switc hing Cen ter (MSC), whic h connects the BSs and bac kb one comm unication infrastructure (suc h as Public Switc hed T elephone Net w ork (PSTN)). Therefore, w e assume that an MSC, an SSP and a VLR database are asso ciated together to serv e an RA. The conguration ma y v ary in practice; ho w ev er, the assumption is used only for p erformance analysis. Since w e do not

PAGE 21

10 deal with the con ten t of the messages, w e assume that the message sizes are equal for all signaling transactions. 2.3 Basic Mobilit y Managemen t Pro cedures of IS-41/GSM MAP In the follo wing sections, w e will compare the p erformances of the new sc hemes with that of the basic IS-41/GSM MAP sc heme. The new sc hemes can minimize the o v erall signaling trac burden b y revising the curren t proto cols. So w e need to dene the basic sc heme rst. T o facilitate the presen tation, the follo wing t w o op erations are dened: MO VE : the user mo v emen t from one RA to another; FIND : determination of the RA where the user is curren tly in. W e call the MO VE and FIND op erations used in curren t mobilit y managemen t standards suc h as IS-41 or GSM MAP the BasicMO VE and BasicFIND W e presen t the procedures in the follo wing pseudo co de. W e remark that the BasicMO VE and BasicFIND pro cedures w e presen t here are simplications of those in the standards; ho w ev er, suc h simplications do capture the ma jor in teractions b et w een the HLR and VLR databases relev an t to our comparativ e study BasicMO VE() f The mobile terminal dete cts that it is in a new r e gistr ation ar e a; The mobile terminal sends a r e gistr ation message to the new VLR; The new VLR sends a r e gistr ation message to the user's HLR; The HLR sends a r e gistr ation c anc el lation message to the old VLR; The old VLR sends a c anc el lation c onrmation message to the HLR; The HLR sends a r e gistr ation c onrmation message to the new VLR; g BasicFIND() f Cal l to a user is dete cte d at the lo c al switch; if the c al le d p arty is in the same RA, then r eturn; Switch queries the c al le d p arty's HLR; HLR queries the c al le d p arty's curr ent VLR V ; VLR V r eturns the c al le d p arty's lo c ation to HLR; HLR r eturns the lo c ation to the c al ling p arty; g

PAGE 22

11 i=0j=0 i=1j=K2i=K1-1 j=K2-1 i=0j=0 i=1j=K2i=1j=1gRA i=2j=K2 . . . REGNOT SSP/MSC/VLR HLR REGPTR_L1 REGPTR_L1 REGPTR_L2 regnot REGPTR_L1 REGPTR_L2 regptr_l1 regptr_l1 regptr_l2 regptr_l1 regptr_l2 . . . RAaRAbRAcRAdRAeRAf Figure 2.2: The TwoL evelFwdMO VE pro cedures 2.4 Tw o-Lev el P oin ter F orw arding Sc heme 2.4.1 Mobilit y Managemen t Pro cedures The t w o-lev el p oin ter forw arding sc heme mo dies the basic pro cedures as follo ws. When a user mo v es from one RA to another, it informs the switc h (and the VLR) at the new RA ab out the old RA. It also informs the new RA ab out the previous MA it w as registered. The switc h at the new RA determines whether to in v ok e the BasicMO VE or the TwoL evelFwdMO VE strategy In TwoL evelFwdMO VE the new VLR exc hanges messages with the old VLR or the old MA to set up a forw arding p oin ter from an old VLR to the new VLR. If a p oin ter is set up from the previous MA, the new VLR is selected as the curren t MA. The TwoL evelFwdMO VE pro cedures do not in v olv e the user's HLR. Fig. 2.2 sho ws the Two-L evel F orwar d MO VE pro cedures with lev el 1 p oin ters c hain threshold limited to 3. Assume that a user mo v es from R A a to R A g (these RAs are not necessary to b e adjacen t) and R A a is the user's MA. When the user lea v es R A a but b efore he en ters R A b the user informs the new VLRs and the lev el 2 p oin ters are built from the old VLR to the new VLR. When the user en ters R A b the c hain threshold for lev el 2 p oin ter is reac hed, so R A b is selected as the user's new MA and a lev el 1 p oin ter is set up from the old MA to the new MA. A t the same time, the lev el 2 p oin ter c hain is reset. The similar pro cedures are used at R A c A lev el 1 p oin ter

PAGE 23

12 SCP/HLR RSTP LSTP SSP/VLR (1) (4) (2) (3) L1 forwarding pointers L1 forwarding pointers L2 forwarding pointers incoming call Figure 2.3: The TwoL evelFwdFIND pro cedures is set up from R A b to R A c and the VLR in R A c is the user's new MA. As the user k eeps mo ving, in R A e the threshold for lev el 2 p oin ter c hain is reac hed again, while this time the threshold of the lev el 1 p oin ter c hain is reac hed to o. Instead of exc hanging information with the previous MA, the BasicMO VE() is in v ok ed. The HLR is up dated with the user curren t lo cation. The messages REGPTR L1 and REGPTR L2 are messages from the new VLR to the old VLR sp ecifying that a lev el 1 or lev el 2 forw arding p oin ter is to b e set up; messages regptr l1 and regptr l2 are the conrmations from the old VLR (or MA). In this gure, the VLRs in R A a R A b R A e and R A f are selected as the user's MAs. The TwolevelFwdFIND pro cedures are in v ok ed for the subsequen t calls to the user from some other switc hes. The user's HLR is queried rst as in the basic strategy and a p oin ter to the user's p oten tially outdated MA is obtained. The p oin ter c hain is follo w ed to nd out the user's curren t lo cation (see Fig. 2.3). As w e can see, in the t w o-lev el p oin ter forw arding sc heme, the c hain length can b e longer than that in the basic p oin ter forw arding sc heme without increasing the Find p enalt y signican tly The previous study [23] sho ws that more sa ving can b e obtained with a longer c hain. Ho w ev er, the p oin ter c hain length is limited b y the dela y restriction requiremen t. By appropriately tuning the t w o thresholds in our sc hemes, w e can mitigate the signaling cost without to o m uc h increase of call setup dela y The t w o-lev el p oin ter forw arding pro cedures can b e describ ed b y the follo wing pseudo co de (w e use the shared global v ariables i and j in the pseudo co de).

PAGE 24

13 TwoL evelFwdMO VE() f /*Initial ly, i; j ar e 0*/ if( j < K 2 and i < K 1 ) f The user r e gisters at the new RA/VLR, p assing the ID of the former RA/VLR and MA; The new VLR der e gisters the user at old VLR; The old VLR sends A CK and the user's servic e pr ole to the new VLR; j := j + 1; g else if( j > = K 2 and i < K 1 ) f The user r e gisters at the new RA/VLR(MA), p assing the ID of the former RA/VLR and MA; The new VLR der e gisters the user at the old MA/VLR; The old MA/VLR sends A CK and the user's servic e pr ole to the new MA/VLR; i := i + 1; j := 0; g elsef BasicMO VE(); i := 0; j := 0; gg TwoL evelFwdFIND() f A c al l to the PCS user is dete cte d at a lo c al switch; if (the c al le d p arty is in the same RA) r eturn; The lo c al switch queries the c al le d p arty's HLR; HLR queries V 0 /MA; While(Querie d VLR is not the c al le d p arty's curr ent VLR); VLR queries the next VLR in the p ointer chain; /*Now the c al le d p arty's actual VLR has b e en found*/ i := 0; j := 0; The c al le d p arty's curr ent VLR sends the user lo c ation to HLR; HLR sends the user lo c ation to c al ling p arty's switch; g

PAGE 25

14 2.4.2 Cost F unctions and P erformance Analysis In this section, w e dev elop an analytic mo del and study the p erformance of the t w o-lev el p oin ter forw arding strategy based on dieren t parameters for dieren t classes of users. W e c haracterize the classes of users according to their call-to-mobilit y ratio ( C M R ). The C M R of a user is dened as the exp ected n um b er of calls to a user during the p erio d that the user visits an RA (notice that the C M R is dened here in terms of the calls receiv ed b y a particular user, not calls originated from the user). If calls are r e c eive d b y the user at a mean rate and the time the user resides in a giv en RA has a mean 1 = then the C M R denoted as is giv en b y = =: (2.1) In order to mak e comparison of the costs, w e need to analyze the basic pro cedures rst. Assume that a user crosses m ultiple RAs b et w een t w o consecutiv e calls. If the basic user lo cation strategy is used, the user's HLR is up dated ev ery time the user mo v es to a new RA. If the t w o-lev el p oin ter forw arding strategy is used, the HLR is up dated only ev ery K 1 K 2 mo v es ( K 1 and K 2 are the lev el 1 and lev el 2 p oin ter c hain length threshold resp ectiv ely), while forw arding p oin ters are set up for all the other mo v es. W e dene C B and C F as the total costs for main taining the lo cation information (lo cation up dating) and lo cating the user (lo cation tracing) b et w een t w o consecutiv e calls for the basic strategy and the t w o-lev el forw arding strategy resp ectiv ely The follo wing notations will b e used in our analysis. m : the cost of a single in v o cation of BasicMO VE ; M : the total cost of all the BasicMO VESs b et w een t w o consecutiv e calls; F : the cost of a single BasicFIND ; M 0 : the exp ected cost of all TwoL evelFwdMO VEs b et w een t w o consecutiv e calls; F 0 : the a v erage cost of the TwoL evelFwdFIND ; S 1 : the cost of setting up a forw arding p oin ter (lev el 1) b et w een MAs during a TwoL evelFwdMO VE ; S 2 : the cost of setting up a forw arding p oin ter (lev el 2) b et w een VLRs during a TwoL evelFwdMO VE ;

PAGE 26

15 T 1 : the cost of tra v ersing a forw arding p oin ter (lev el 1) b et w een MAs during a TwoL evelFwdFIND ; T 2 : the cost of tra v ersing a forw arding p oin ter (lev el 2) b et w een VLRs during a TwoL evelFwdFIND ; K 1 : the threshold of lev el 1 p oin ter c hain; K 2 : the threshold of lev el 2 p oin ter c hain; ( i ) : the probabilit y that there are i RA crossings b et w een t w o consecutiv e calls. Then, w e ha v e C B = M + F = m=p + F : (2.2) C F = M 0 + F 0 : (2.3) No w, w e can deriv e form ulas for M 0 and F 0 as follo ws. Supp ose a user crosses i RA b oundaries b et w een t w o consecutiv e calls. The HLR is up dated b i K 1 K 2 c times. There are also b i K 2 c b i K 1 K 2 c lev el 1 p oin ter creations (ev ery K 2 mo v es ma y require a lev el 1 p oin ter creation but sometimes the HLR is up dated and lev el 1 p oin ter is not set up). The lev el 2 p oin ters are created for all the rest i b i K 2 c mo v es. Th us, w e obtain M 0 = 1 X i =0 ( i K 1 K 2 m + i K 2 i K 1 K 2 S 1 + i i K 2 S 2 ) ( i ) : (2.4) The cost of F 0 is deriv ed as follo ws. After the last BasicMove op erations (if an y), the user tra v erses $ i j i K 1 K 2 k K 1 K 2 K 2 % lev el 1 p oin ters and i i K 1 K 2 K 1 K 2 $ i j i K 1 K 2 k K 1 K 2 K 2 % K 2 lev el 2 p oin ters. Th us, w e obtain F 0 = F + 1 X i =0 ($ i j i K 1 K 2 k K 1 K 2 K 2 % T 1 + i i K 1 K 2 K 1 K 2 $ i j i K 1 K 2 k K 1 K 2 K 2 % K 2 T 2 ) ( i ) : (2.5) In order to ev aluate ( i ), w e mak e the follo wing assumptions. 1. The call arriv als to a user form a P oisson pro cess with arriv al rate

PAGE 27

16 2. The residence time of a user at a registration area is a random v ariable with a general densit y function f m ( t ) and the Laplace transform is f m ( s ) = Z 1 t =0 f m ( t ) e st dt: The exp ected residence time of a user in an RA is 1 = W e denote g = f m ( ) for con v enience. With these assumptions, (2.4) and (2.5) can b e deriv ed as follo ws. M 0 = 1 X i =0 ( i K 1 K 2 m + i K 2 i K 1 K 2 S 1 + i i K 2 S 2 ) ( i ) = 1 X i =0 iS 2 ( i ) | {z } X + 1 X i =0 i K 2 ( S 1 S 2 ) ( i ) | {z } Y + 1 X i =0 i K 1 K 2 ( m S 1 ) ( i ) | {z } Z : X can b e simplied from the denition of ( i ), X = S 2 1 X i =0 i ( i ) = S 2 : The probabilit y ( i ) can b e expressed as (see [48] for the detailed deriv ation), ( i ) = (1 g ) 2 g i 1 : (2.6) Y = ( S 1 S 2 ) 1 X i =0 i K 2 ( i ) : Let i = j K 2 + k then, ( j K 2 + k ) = (1 g ) 2 g ( g K 2 ) j g k = y z j x k : where y = (1 g ) 2 g ; z = g K 2 ; x = g :

PAGE 28

17 W e can rewrite, Y = y ( S 1 S 2 ) 1 X j =0 K 2 1 X k =0 j z j x k = y ( S 1 S 2 )(1 x K 2 ) 1 x ( 1 X j =0 j z j ) = y z ( S 1 S 2 )(1 x K 2 ) (1 x )(1 z ) 2 = (1 g )( S 1 S 2 ) g K 2 1 (1 g K 2 ) : Similarly Z = ( m S 1 ) 1 X i =0 i K 1 K 2 ( i ) = (1 g )( m S 1 ) g K 1 K 2 1 (1 g K 1 K 2 ) : So, w e obtain M 0 = S 2 + (1 g ) g K 2 1 ( S 1 S 2 ) (1 g K 2 ) + (1 g ) g K 1 K 2 1 ( m S 1 ) (1 g K 1 K 2 ) : (2.7) W e can deriv e F 0 in a similar w a y F 0 = F + 1 X i =0 ($ i j i K 1 K 2 k K 1 K 2 K 2 % T 1 + i i K 1 K 2 K 1 K 2 $ i j i K 1 K 2 k K 1 K 2 K 2 % K 2 T 2 ) ( i ) = F + ( T 1 K 2 T 2 ) 1 X i =0 $ i j i K 1 K 2 k K 1 K 2 K 2 % ( i ) | {z } U + T 2 1 X i =0 i i K 1 K 2 K 1 K 2 ( i ) | {z } V : Let i = j K 1 K 2 + k then, ( j K 1 K 2 + k ) = (1 g ) 2 g ( g K 1 K 2 ) j g k = y z j x k ; where y = (1 g ) 2 g ; z = g K 1 K 2 ; x = g :

PAGE 29

18 U = ( T 1 K 2 T 2 ) 1 X i =0 $ i j i K 1 K 2 k K 1 K 2 K 2 % ( i ) = ( T 1 K 2 T 2 ) 1 X j =0 K 1 K 2 1 X k =0 k K 2 ( j K 1 K 2 + k ) = ( T 1 K 2 T 2 ) 1 X j =0 K 1 K 2 1 X k =0 k K 2 y z j x k = ( T 1 K 2 T 2 ) y 1 X j =0 z j K 1 K 2 1 X k =0 k K 2 x k = ( T 1 K 2 T 2 ) y 1 X j =0 z j K 1 1 X n =0 K 2 1 X m =0 nx nK 2 + m = ( T 1 K 2 T 2 )(1 g )[ g K 2 K 1 g K 1 K 2 + ( K 1 1) g ( K 1 +1) K 2 ] pg (1 g K 1 K 2 )(1 g K 2 ) ; V = T 2 1 X i =0 i i K 1 K 2 K 1 K 2 ( i ) = T 2 1 X j =0 K 1 K 2 1 X k =0 k ( j K 1 K 2 + k ) = y T 2 1 X j =0 z j K 1 K 2 1 X k =0 k x k = [1 K 1 K 2 g K 1 K 2 1 + ( K 1 K 2 1) g K 1 K 2 ] T 2 (1 g K 1 K 2 ) : So w e ha v e F 0 = F + T 2 [1 K 1 K 2 g K 1 K 2 1 + ( K 1 K 2 1) g K 1 K 2 ] (1 g K 1 K 2 ) + ( T 1 K 2 T 2 )(1 g )[ g K 2 K 1 g K 1 K 2 + ( K 1 1) g ( K 1 +1) K 2 ] pg (1 g K 1 K 2 )(1 g K 2 ) : (2.8) F or demonstration purp oses, w e assume that the RA residence time of a user is Gamma distributed with mean 1 = The reason that Gamma distribution is selected is its rexibilit y in setting v arious parameters and can b e used to t the rst t w o momen ts of the eld data. The Laplace transform of a Gamma distribution is f m ( s ) = r s + r r ;

PAGE 30

19 th us, w e ha v e, g = f m ( ) = r + r r = r + r r : (2.9) In particular, when r = 1, w e ha v e an exp onen tial distribution for the RA residence time. 2.4.3 P erformance Analysis under Exp onen tial RA Residence Time W e rst consider the situation when the RA residence time is exp onen tially distributed. By setting r = 1 in (2.9), w e ha v e g = 1 1 + : Then (2.7) and (2.8) can b e rewritten as M 0 = S 2 + S 1 S 2 (1 + ) K 2 1 + m S 1 (1 + ) K 1 K 2 1 ; (2.10) F 0 = F + T 2 T 2 K 1 K 2 (1 + ) K 1 K 2 1 + ( T 1 K 2 T 2 )[(1 + ) K 1 K 2 K 1 (1 + ) K 2 + K 1 1] [(1 + ) K 1 K 2 1][(1 + ) K 2 1] : (2.11) F rom (2.10), (2.11) and (2.3), w e obtain C F = F + T 2 + S 2 + S 1 S 2 (1 + ) K 2 1 + m S 1 T 2 K 1 K 2 (1 + ) K 1 K 2 1 + ( T 1 K 2 T 2 )[(1 + ) K 1 K 2 K 1 (1 + ) K 2 + K 1 1] [(1 + ) K 1 K 2 1][(1 + ) K 2 1] : (2.12) W e notice that up dating the HLR and p erforming a BasicFIND in v olv e the same n um b er of messages b et w een HLR and VLR databases, so w e can c ho ose m = F Without loss of generalit y w e can normalize m = 1. W e also assume that the cost of setting up a forw arding p oin ter is ab out t wice the cost of tra v ersing it, since t wice as man y messages are in v olv ed, i.e., w e set S 1 = 2 T 1 and S 2 = 2 T 2 W e consider S 2 = with < 1. Since the lev el 1 p oin ter is more exp ensiv e than lev el 2 p oin ter in terms of setup cost, w e can assume S 1 = K S 2 with K 1. It is reasonable to assume that S 1 < 1 to o. W e will see later, ho w ev er, that the t w o-lev el forw arding strategy can also p erform w ell ev en with S 1 1. F rom (2.2), (2.10), (2.11) and (2.12), w e obtain, C B = 1 + 1 ; (2.13)

PAGE 31

20 0 1 2 0.3 0.32 0.34 0.36 0.38 0.4 0.42 p(CMR) (a) The MOVE cost M'/M K1=2;K2=4K1=2;K2=6K1=4;K2=4K1=4;K2=6 0 1 2 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 p(CMR) (b) The FIND cost F'/F K1=2;K2=4K1=2;K2=6K1=4;K2=4K1=4;K2=6 0 1 2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 p(CMR) (c) The net cost CF/CB K1=2;K2=4K1=2;K2=6K1=4;K2=4K1=4;K2=6 Figure 2.4: Relativ e MO VE and FIND costs of forw arding with = 0 : 3 ; K = 1 : 5 M 0 M = + ( K 1) (1 + ) K 2 1 + (1 K ) (1 + ) K 1 K 2 1 ; (2.14) F 0 F = 1 + 2 K 1 K 2 2[(1 + ) K 1 K 2 1] + ( K K 2 )[(1 + ) K 1 K 2 K 1 (1 + ) K 2 + K 1 1] 2[(1 + ) K 1 K 2 1][(1 + ) K 2 1] ; (2.15) C F C B = 1 + 1 + 3 2 + ( K 1) (1 + ) K 2 1 + 1 ( K + 1 2 K 1 K 2 ) (1 + ) K 1 K 2 1 + ( K K 2 )[(1 + ) K 1 K 2 K 1 (1 + ) K 2 + K 1 1] 2[(1 + ) K 1 K 2 1][(1 + ) K 2 1] : (2.16) In Fig. 2.4, 2.5 and 2.6, w e plot the costs as functions of C M R for v arious v alues of K 1 ; K 2 ; K and Fig. 2.4(a) sho ws that under certain conditions ( = 0 : 3 ; K = 1 : 5), the t w o-lev el forw arding sc heme can result in 60% 70% reductions in lo cation up date cost comparing with the basic strategy Ho w ev er, the Fig. 2.4(b) indicates that the FIND cost of the t w o-lev el forw arding sc heme is higher than the basic strategy The reason is that the call setup messages for the user need tra v erse the p oin ter c hain to nd the user's curren t lo cation. Ho w ev er, as w e observ e in Fig. 2.4(c), the t w o-lev el forw arding strategy can result in 20% 60% reduction of the total cost. If w e study the plots carefully w e can observ e that b oth the relativ e MO VE an FIND costs are decreasing functions of ( C M R ). When is small, the user crosses RAs more frequen tly The p oin ters are needed to b e set up and a

PAGE 32

21 0 1 2 0.3 0.35 0.4 0.45 0.5 0.55 p(CMR) (a) The MOVE cost M'/M K1=2;K2=4K1=2;K2=6K1=4;K2=4K1=4;K2=6 0 1 2 1 1.5 2 p(CMR) (b) The FIND cost F'/F K1=2;K2=4K1=2;K2=6K1=4;K2=4K1=4;K2=6 0 1 2 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 p(CMR) (c) The net cost C F /C B K1=2;K2=4K1=2;K2=6K1=4;K2=4K1=4;K2=6 Figure 2.5: Relativ e MO VE and FIND costs of forw arding with = 0 : 3 ; K = 4 long c hain of p oin ters ha v e to b e tra v ersed, leading to the high setup cost. The impro v emen t of the total cost increases when the decreases b ecause most MO VEs do not result in HLR up dates but p oin ter creations. Consider the Fig. 2.4(a) again, w e can ha v e more sa ving in the MO VEs with longer p oin ter c hain b ecause more up dates to HLR can b e substituted with p oin ter creations. Ho w ev er, long p oin ter c hain increases the FIND p enalt y at the same time (see Fig. 2.4(b)). An adv an tage of the t w o-lev el forw arding strategy is that it can ha v e long p oin ter c hain without increasing the dela y p enalt y signican tly b ecause the p oin ter c hain can b e shortened b y the lev el 1 p oin ters b et w een MAs. Under the assumed conditions, the maxim um p oin ter c hain length can increase from 8 to 24 with only 30% FIND p enalt y increase. The Fig. 2.5 sho ws the relativ e cost curv es when K increases from 1 : 5 to 4. As w e can see, ev en in this case, the cost of setting up a lev el 1 p oin ter exceeds that of up dating HLR, there is only sligh t increase of the total cost. The MO VE and FIND costs b oth increase b ecause the cost of setting up and tra v ersing lev el 1 p oin ters c hain increases. Since lev el 1 p oin ter is built up only when the lev el 2 p oin ter c hain threshold is reac hed and the n um b er of lev el 2 p oin ters is dominan t, therefore the t w o-lev el forw arding strategy is not sensitiv e to the v ariation of K The Fig. 2.6 indicates that the lev el 2 p oin ter op eration cost has more eect on the system p erformance. In Fig. 2.6, is increased from 0 : 3 to 0 : 6. The MO VE,

PAGE 33

22 0 1 2 0.6 0.61 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 p(CMR) (a) The MOVE cost M'/M K 1 =2;K 2 =4 K 1 =2;K 2 =6 K 1 =4;K 2 =4 K 1 =4;K 2 =6 0 1 2 1 1.5 2 2.5 p(CMR) (b) The FIND cost F'/F K 1 =2;K 2 =4 K 1 =2;K 2 =6 K 1 =4;K 2 =4 K 1 =4;K 2 =6 0 1 2 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 p(CMR) (c) The net cost C F /C B K 1 =2;K 2 =4 K 1 =2;K 2 =6 K 1 =4;K 2 =4 K 1 =4;K 2 =6 Figure 2.6: Relativ e MO VE and FIND costs of forw arding with = 0 : 6 ; K = 1 : 5 FIND and the net costs all increase. Finally w e can observ e that for small increasing p oin ter c hain length reduces the cost of t w o-lev el forw arding sc heme b ecause the p oin ter op erations are c heap er. 2.4.4 Sensitivit y Analysis to the Residence Time W e no w in v estigate the sensitivit y of the p erformance costs and b enets of the t w olev el forw arding sc heme to the v ariance of the user's mobilit y patterns. W e assume that the call arriv als to a user form a P oisson pro cess, and the RA residence time has a Gamma distribution. F or a Gamma distribution, the v ariance is V = 2 r In other w ords, a large r implies a small v ariance. Fig. 2.7 sho ws the eect of r on M 0 = M ; F 0 =F and C F =C B In Fig. 2.7, w e observ e that the increase of the v ariance of RA residence time (smaller r ) causes the increase of M 0 = M but the decrease of F 0 =F ; the net eect to C F =C B is not signican t. Consider M 0 = M for r < 1 comparing with the case when r = 1, for a giv en > 0 (see Fig. 2.7(a)), the large v ariance of RA residence time implies high v ariation of the RA b oundary crossing patterns. If the user crosses man y RAs, a longer p oin ter c hain will b e created. When the c hain limit K 1 K 2 reac hes, the HLR will b e up dated, resulting in the increase of M 0 On the other hand, if few er b oundaries are crossed, only shorter p oin ter c hains will b e set up, the p oin ter creation/tracing cost will b e sa v ed. The net eect is an

PAGE 34

23 0 1 2 0.3 0.31 0.32 0.33 0.34 0.35 0.36 0.37 p(CMR) (a) The MOVE cost M'/M r=100 r=1 r=0.01 0 1 2 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 p(CMR) (b) The FIND cost F'/F r=100 r=1 r=0.01 0 1 2 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 p(CMR) (c) The net cost CF/CB r=100 r=1 r=0.01 Figure 2.7: The eect of v ariance in residence time ( r ) with = 0 : 3 ; K = 1 : 5 ; K 1 = 4 and K 2 = 4 increase in M 0 No w, consider the eect of v ariance in the RA residence time to the F 0 =F (See Fig. 2.7(b)). When the v ariance is high, the n um b er of RA b oundaries the user crosses b et w een t w o consecutiv e calls will v ary greatly When the n um b er is small, the FIND cost is reduced; when the n um b er is large, the p oin ter c hain could b e shortened. The net eect is a signican t impro v emen t in F 0 =F The net eect of the v ariance of the residence time on total cost ratio C F =C B is not signican t for lo w C M R (see Fig. 2.7(c)). 2.5 POFLA Sc heme W e will in tro duce another p oin ter forw arding based sc heme{P oin ter F orw arding Based Lo cal Anc horing Sc heme (POFLA) in this section. 2.5.1 The Sc heme Ov erview In the POFLA sc heme, the basic lo cation up date and call deliv ery pro cedures are mo died to ac hiev e b etter p erformance. A VLR serving a user is selected as the MA for that user and ma y c hange during the user's mo v emen t. Since the selection of MAs for the users is based on their lo cations, hence the signaling trac can b e distributed ev enly among the net w ork. This can a v oid the b ottlenec k eect normally exp erienced b y the HLR. The basic lo cation up date pro cedure is mo died as follo ws: ev ery time a user en ters a new RA serv ed b y a dieren t VLR, the mobile terminal registers to the new VLR and

PAGE 35

24 MA MA RA 1 RA 2 RA 3 RA 4 RA 5 RA 6 RA 7 RA 8 MSC8 VLR8 MSC7 VLR7 MSC6 VLR6 MSC5 VLR5 MSC4 VLR4 MSC3 VLR3 MSC2 VLR2 MSC1 VLR1 HLR L Pointer L Pointers L Pointers L Pointers H Pointer H Pointer Figure 2.8: POFLA strategy pro cedures informs the new VLR ab out the old VLR and MA. The old VLR and MA ma y b e the same VLR the user curren tly visited. The VLR at the new RA determines what to do based on the information from the mobile lo cation up date b eha viors. The new VLR has three options: it can request the old VLR to set up a p oin ter to itself, whic h is called L ow L evel Pointer (or L P oin ter) in our sc heme; it can up date the MA and request to set up a p oin ter from the MA to itself, this is called High L evel Pointer (or H P oin ter); and it can also decide to up date the user's new lo cation to the HLR directly and it itself b ecomes the new MA. Fig. 2.8 sho ws the lo cation up date and call deliv ery pro cedures in the POFLA sc heme with the H p oin ter n um b er limit b eing three. Assume a mobile user mo v es from R A 1 to R A 8 (these RAs are not necessary to b e adjacen t) and V LR 1 is the user's curren t MA. A t the b eginning, the user is in R A 1 and V LR 1 is the user's curren t serving VLR. The V LR 1 is selected as the user's curren t MA b ecause either the user just receiv es an incoming call in R A 1 or the V LR 1 just up dates the user's new lo cation to the HLR. When the user lea v es R A 1 but b efore en ters R A 3 the mobile terminal informs the new VLR and a p oin ter c hain consisting of L p oin ters is set up just as in the p er-user forw arding sc heme [23]. When the user en ters R A 3 the c hain threshold for L p oin ters is reac hed. In this situation, the V LR 3 will up date the user's new lo cation to the curren t MA, i.e. V LR 1 A t the same time, the L p oin ter c hain is reset. The same pro cedure is used in V LR 5 and the previous H p oin ter is reset. If the user k eeps mo ving, in R A 7 the threshold for L p oin ter c hain is reac hed again. This time, the limit of the H p oin ter n um b er is reac hed to o. Instead of exc hanging information with the previous MA and setting up a new H p oin ter, the V LR 7 will up date the user's lo cation to the HLR directly and V LR 7 is selected as the new MA for that the user. The reason of up dating

PAGE 36

25 the HLR instead of the MA is that the cost of setting up and tra v ersing the p oin ter c hain b et w een MA and curren t serving VLR ma y b e costly when the user is far a w a y from the MA and the connection setup dela y for an incoming call ma y b e in tolerable. If an incoming call arriv es b efore the mobile user c hanges his or her MA, the curren t serving VLR b ecomes the user's curren t MA b ecause the HLR has the kno wledge of the user's curren t lo cation after the connection setup and it is not necessary to go through the p oin ter c hain to lo cate the user again for the future service deliv eries. The call deliv ery pro cedure in the POFLA sc heme is straigh tforw ard. When the subsequen t calls are initiated from some other switc hes to the user, the user's HLR is queried rst as in the basic pro cedure and a p oin ter to the user's p oten tially outdated MA is obtained. The p oin ter c hain is follo w ed to nd the user's curren t lo cation. F or example, in Fig. 2.8, if the user is in R A 6 when a call arriv es. The user can b e reac hed b y follo wing the H p oin ter rst and L p oin ter c hain second. After the connection setup, the curren t serving VLR b ecomes the user's new MA. As w e can see, in the POFLA sc heme, the c hain length can b e longer than that in the basic p oin ter forw arding sc heme without increasing the connection setup dela y signican tly In section 2.5.3, w e can see that the dela y for the POFLA sc heme is m uc h less than that for the p er-user forw arding sc heme. Our study also sho ws, under some assumption, the POFLA sc heme p erforms b etter than the static and dynamic lo cal anc hor sc hemes prop osed b y Ho and Akyildiz [22]. Although in section 2.5.3, the t w o-lev el p oin ter forw arding sc heme [45] can generate similar results as that of the POFLA sc heme, the new sc heme is simpler to implemen t than the t w o-lev el p oin ter forw arding sc heme in practical systems. 2.5.2 System Mo del and Cost F unctions In this section, w e dev elop an analytic mo del to deriv e the cost functions and compare the p erformance of the POFLA sc heme with IS-41/GSM MAP p oin ter forw arding, lo cal anc hor and t w o-lev el p oin ter forw arding sc hemes. The mobile users in a PCS can b e c haracterized b y their c al l-to-mobility r atios (CMRs). All the ab o v e sc hemes are ev aluated under dierence CMRs in this c hapter. W e observ e that a mobile needs to up date its lo cation only when the mobile do es not engage an y comm unication with other users. Hence w e can only need to compare the signaling trac in the

PAGE 37

26 time in terv al b et w een call services (i.e., the in terv al b et w een the end of the curren t call and the b eginning of the next call, whic h is called inter-servic e time in [49, 50]). Assume that a mobile crosses a n um b er of RAs during in ter-service time. By ignoring the busy-line eect, the in ter-service time can b e appro ximated b y the in ter-arriv al time of calls to the mobile. If the basic user lo cation up date sc heme (IS-41) is used, the user's HLR will b e up dated ev ery time the user mo v es to a new RA. In the POFLA sc heme, the HLR is up dated only ev ery K 1 K 2 mo v es ( K 1 and K 2 are the L p oin ter c hain threshold and H p oin ter n um b er limit, resp ectiv ely), and p oin ters are set up for all other mo v es. W e dene C 0 to b e the total costs of up dating the lo cation information (lo cation up date) and trac king the user (call deliv ery) during the in ter-service time for the the POFLA strategies. F or con v enience, w e list all notations used in our analysis as follo ws: K 1 : the threshold for the L p oin ter c hain; K 2 : the limit of H p oin ter n um b er (i.e., ev ery K 2 up dates to an MA will result in the c hange of a new MA); m : the a v erage cost of lo cation up date to the HLR; M : the total lo cation up date cost during the in ter-service time in the IS-41 sc heme; F : the total cost of call deliv ery in the IS-41 sc heme; M 0 : the total lo cation up date cost in the POFLA sc heme during the in ter-service time; F 0 : the total call deliv ery cost in the POFLA sc heme; S 1 : the p oin ter setup cost of an L p oin ter; S 2 ;j : the p oin ter setup cost of the j th H p oin ter; T 1 : the cost of tra v ersing an L p oin ter b et w een t w o adjacen t VLRs; T 2 ;j : the cost of tra v ersing the j th H p oin ter; ( i ) : the probabilit y that there are i RA crossings during the in ter-service time; P : the pro cessing cost of setting up a p oin ter (H or L); G : the signaling cost of setting up an L p oin ter; : the co ecien t of signaling cost for an H p oin ter ( 1); : the user Call-to-mobilit y ratio.

PAGE 38

27 Then, w e can express the total costs during the in ter-service time for the POFLA sc heme as follo ws: C 0 = M 0 + F 0 : (2.17) Since p oin ters are set up in the POFLA sc heme, w e need dene the p oin ter setup and tra v ersing costs for further analysis. Ev ery time a p oin ter is set up, the signaling messages will b e transmitted bac k and forth; F or p oin ter tra v ersing, the signaling message is transmitted only in one direction. So w e dene the costs of p oin ter setup and tra v ersing for the L p oin ter as S 1 and T 1 resp ectiv ely: S 1 = G + P ; (2.18) T 1 = 1 2 G + P : (2.19) The costs of H p oin ters are not xed b ecause the length of the H p oin ter c hanges with the user's mobilit y In this c hapter, w e express the costs for H p oin ters as follo ws: S 2 ;j = 8><>: 0 if j = 0 j G + P Otherwise (2.20) T 2 ;j = 8><>: 0 if j = 0 1 2 j G + P Otherwise (2.21) where the j means the setup cost or the tra v ersing cost for the j th H p oin ter. No w, w e can deriv e the form ulas for M 0 and F 0 as follo ws: supp ose that a user crosses i RA b oundaries during the in ter-service time. The HLR is up dated b i K 1 K 2 c times. If w e call the summation from the 0 th to the ( K 2 1) th H p oin ter setup cost, P K 2 1 j =0 S 2 ;j the MA up date cost, then, there are b i K 1 K 2 c times of suc h MA up date costs that w ould incur during the in ter-service time with i RA crossings. In addition, there are b i b i K 1 K 2 c K 1 K 2 K 1 c H p oin ter setups and i b i K 1 c L p oin ter setups o ccurred in the remaining i RA crossings. Th us, w e can obtain M 0 = 1 X i =0 fb i K 1 K 2 c m + ( i b i K 1 c ) S 1 + b i K 1 K 2 c ( K 2 1 X j =0 S 2 ;j ) + b i b i K 1 K 2 c K 1 K 2 K 1 c X j =0 S 2 ;j g ( i ) : (2.22)

PAGE 39

28 The cost F 0 can b e deriv ed straigh tforw ard. In order to reac h the user's curren t lo cation, the signaling message is sen t to the MA and then tra v els through one H p oin ter (if an y) and i b i K 1 c K 1 L p oin ters b efore reac hing the curren t lo cation. So, w e ha v e F 0 = F + 1 X i =0 f T 2 ; + ( i b i K 1 c K 1 ) T 1 g ( i ) ; (2.23) where = b i b i K 1 K 2 c K 1 K 2 K 1 c : Notice that w e can easily obtain K 2 1 X j =0 S 2 ;j = K 2 ( K 2 1) G + 2( K 2 1) P 2 ; and b i b i K 1 K 2 c K 1 K 2 K 1 c X j =0 S 2 ;j = ( b i b i K 1 K 2 c K 1 K 2 K 1 c + 1) b i b i K 1 K 2 c K 1 K 2 K 1 c G + 2 b i b i K 1 K 2 c K 1 K 2 K 1 c P 2 : So (2.22) can b e rewritten as follo w, M 0 = S 1 1 X i =0 i ( i ) | {z } M 1 S 1 1 X i =0 b i K 1 c ( i ) | {z } M 2 + ( K 2 1) K 2 G + 2( K 2 1) P + 2 m 2 1 X i =0 b i K 1 K 2 c ( i ) | {z } M 3 + G 2 1 X i =0 i b i K 1 K 2 c K 1 K 2 K 1 2 ( i ) | {z } M 4 + G + 2 P 2 1 X i =0 i b i K 1 K 2 c K 1 K 2 K 1 ( i ) | {z } M 5 : (2.24) M 1 can b e simplied from the denition of ( i ), M 1 = S 1 = G + P : (2.25) According to (2.6) and applying v ariable substitution i = j K 1 + k then w e obtain ( j K 1 + k ) = (1 g ) 2 g ( g K 1 ) j g k = y z j x k ; where y = (1 g ) 2 g ; z = g K 1 ; x = g :

PAGE 40

29 Th us, w e ha v e M 2 = y S 1 1 X j =0 K 1 1 X k =0 j z j x k = y S 1 (1 x K 1 ) 1 x ( 1 X j =0 j z j ) = (1 g )( G + P ) g K 1 1 (1 g K 1 ) : (2.26) Similarly w e can obtain M 3 b y using substitution i = j K 1 K 2 + k then ( j K 1 K 2 + k ) = (1 g ) 2 g ( g K 1 K 2 ) j g k = y z j x k : where y = (1 g ) 2 g ; z = g K 1 K 2 ; x = g : M 3 = ( K 2 1) K 2 G + 2( K 2 1) P + 2 m 2 1 X i =0 b i K 1 K 2 c ( i ) = ( K 2 1) K 2 G + 2( K 2 1) P + 2 m 2 (1 g ) g K 1 K 2 1 (1 g K 1 K 2 ) : (2.27) In a similar manner, w e can also obtain M 4 and M 5 as follo ws: M 4 = G 2 1 X j =0 K 1 K 2 1 X k =0 b k K 1 c 2 ( j K 1 K 2 + k ) = G 2 1 X j =0 K 1 K 2 1 X k =0 b k K 1 c 2 y z j x k = G 2 y 1 X j =0 z j ( K 2 1 X n =0 K 1 1 X m =0 n 2 x nK 1 + m ) = G (1 g ) 2 g (1 g K 1 K 2 )(1 g K 1 ) 2 f g K 1 + g 2 K 1 K 2 2 g K 1 K 2 +(2 K 2 2 2 K 2 1) g K 1 ( K 2 +1) ( K 2 1) 2 g K 1 ( K 2 +2) g ; (2.28)

PAGE 41

30 and M 5 = G + 2 P 2 1 X j =0 K 1 K 2 1 X k =0 b k K 1 c ( j K 1 K 2 + k ) = G + 2 P 2 1 X j =0 K 1 K 2 1 X k =0 b k K 1 c y z j x k = G + 2 P 2 y 1 X j =0 z j K 2 1 X n =0 nx nK 1 K 1 1 X m =0 x m = ( G + 2 P )(1 g )[( K 2 1) g K 1 ( K 2 +1) K 2 g K 1 K 2 + g K 1 ] 2 g (1 g K 1 K 2 )(1 g K 1 ) : (2.29) Finally w e nd the expression M 0 = M 1 M 2 + M 3 + M 4 + M 5 If T 2 ; 6 = 0, w e can obtain F 0 in the follo wing fashion F 0 = F + 1 X i =0 f 1 2 b i b i K 1 K 2 c K 1 K 2 K 1 c G + P + ( i b i K 1 c K 1 )( 1 2 G + P ) g ( i ) = F + ( 1 2 G + P ) 1 X i =0 i ( i ) | {z } F 1 ( 1 2 G + P ) K 1 1 X i =0 b i K 1 c ( i ) | {z } F 2 + ( 1 2 G 1 X i =0 b i b i K 1 K 2 c K 1 K 2 K 1 c + P ) ( i ) | {z } F 3 : (2.30) Notice that if w e use the substitution i = j K 1 K 2 + k when k = 0 ; 1 ; ; K 1 1, T 2 ; = 0. So w e can obtain F 0 as follo ws: F 0 = F + G + 2 P 2 [1 K 1 (1 g ) g K 1 1 1 g K 1 ] + P (1 g )( g K 1 1 g K 1 K 2 1 ) (1 g K 1 K 2 ) + G (1 g )[( K 2 1) g K 1 ( K 2 +1) K 2 g K 1 K 2 + g K 1 ] 2 g (1 g K 1 K 2 )(1 g K 1 ) : (2.31) In summary w e obtain Theorem : If the inter-servic e time is exp onential ly distribute d and the RA r esidenc e time is gener al ly distribute d with L aplac e tr ansform f m ( s ) then the total lo c ation up date c ost and

PAGE 42

31 total c ost for c al l delivery ar e given by M 0 = G + P (1 g )( G + P ) g K 1 1 (1 g K 1 ) + ( K 2 1) K 2 G + 2( K 2 1) P + 2 m 2 (1 g ) g K 1 K 2 1 (1 g K 1 K 2 ) + G (1 g ) 2 g (1 g K 1 K 2 )(1 g K 1 ) 2 f g K 1 + g 2 K 1 K 2 2 g K 1 K 2 + (2 K 2 2 2 K 2 1) g K 1 ( K 2 +1) ( K 2 1) 2 g K 1 ( K 2 +2) g + ( G + 2 P )(1 g )[( K 2 1) g K 1 ( K 2 +1) K 2 g K 1 K 2 + g K 1 ] 2 g (1 g K 1 K 2 )(1 g K 1 ) ; F 0 = F + G + 2 P 2 [1 K 1 (1 g ) g K 1 1 1 g K 1 ] + P (1 g )( g K 1 1 g K 1 K 2 1 ) (1 g K 1 K 2 ) + G (1 g )[( K 2 1) g K 1 ( K 2 +1) K 2 g K 1 K 2 + g K 1 ] 2 g (1 g K 1 K 2 )(1 g K 1 ) ; (2.32) wher e g = f m ( ) 2.5.3 P erformance Ev aluations W e consider the situation when the RA residence time is exp onen tially distributed. In our analysis, w e do not address issues regarding the con ten ts of messages and other information transfer whic h ma y o ccur during a call connection setup. F or simplicit y w e assume that the message sizes are equal for all signaling transactions. Since w e only compare the relativ e p erformance of the aforemen tioned sc hemes with the POFLA sc heme under v arious CMRs, the conclusions will not b e aected b y this simplication. Notice that, in the simplied IS-41 or GSM MAP pro cedures, the lo cation up date and call deliv ery in v olv e the same n um b er of messages b et w een HLR and VLR databases, so w e c ho ose m = F Without loss of generalit y w e normalize m = F = 1. G is the signaling transmission cost and P is the pro cessing cost. P usually includes the database transaction costs. The v alues of P and G should b e m uc h less than m or F In our w orks, w e do not assign an y practical meanings to these parameters, they can b e explained as the signaling message trac exc hanged during the lo cation managemen t pro cedures or the time dela y exp erienced in the real systems. In Fig. 2.9, w e plots the relativ e lo cation up date, call deliv ery and net costs of three sc hemes as functions of C M R Here for the POFLA sc heme and the t w o-lev el p oin ter forw arding sc heme, w e assume K 1 = K 2 = 3; for the p er-user forw arding sc heme, the threshold is nine ( K 1 K 2 ). In this gure, w e also assume P = 0 : 05, G = 0 : 1 and = 1 : 5. As w e can see in Fig. 2.9(a), the POFLA sc heme generates higher v alues than the p er-user

PAGE 43

32 10 -2 10 0 10 2 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 (a) Location update cost Call-to-mobility ratio POFLATwo_levelPer-user 10 2 10 0 10 2 1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 (b) Call delivery cost Calltomobility ratio POFLATwo_levelPeruser 10 2 10 0 10 2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 (c) Net cost Calltomobility ratio POFLATwo_levelPeruser Figure 2.9: The relativ e costs for the three sc hemes with P = 0 : 05, G = 0 : 1 and = 1 : 5 forw arding sc heme. It is ob vious b ecause in the latter sc heme, only the L p oin ters are set up while in the POFLA sc heme, a new VLR w ould set up an H p oin ter to the MA when the threshold for L p oin ters is reac hed, whic h costs more than an L p oin ter. F or the t w olev el p oin ter forw arding sc heme, the lev el 2 p oin ter is the L p oin ter and the lev el 1 p oin ter is usually shorter than the H p oin ter [45]. So the lo cation up date cost for the t w o-lev el p oin ter forw arding sc heme is in the middle of them. Although, with the same length of the p oin ter threshold, the p er-user forw arding sc heme can generate less lo cation up date cost, it has the largest call deliv ery cost among the three strategies (see Fig. 2.9(b)). F or some users with small C M R whic h means that the users ha v e higher mobilit y relativ e to call arriv al, the call deliv ery cost for the p er-user forw arding sc heme is m uc h higher than those for the other t w o sc hemes. In practical systems, it could b e em b o died as the dela y the users ha v e to w ait b efore an y connections can b e set up. In the POFLA sc heme, normally few er p oin ters ha v e to b e tra v ersed than the t w o-lev el p oin ter forw arding sc heme b efore a user can b e lo cated, so the POFLA sc heme has the least call deliv ery cost. In Fig. 2.9(b), the p erformance of the POFLA sc heme dose not impro v e m uc h comparing with the t w o-lev el p oin ter sc heme; ho w ev er the POFLA is easier to implemen t in practical systems. Although the three sc hemes p erform dieren tly in lo cation up date and call deliv ery the total net cost for the three sc hemes are similar for high CMRs (Fig. 2.9(c)). In Fig. 2.10, w e increase

PAGE 44

33 10 2 10 0 10 2 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 (a) Location update cost Calltomobility ratio POFLATwo_levelPeruser 10 2 10 0 10 2 1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 (b) Call delivery cost Calltomobility ratio POFLATwo_levelPeruser 10 2 10 0 10 2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 (c) Net cost Calltomobility ratio POFLATwo_levelPeruser Figure 2.10: The relativ e costs for the three sc hemes with P = 0 : 05, G = 0 : 1 and = 3 the signaling transfer co ecien t from 1 : 5 to 3, whic h means that the H p oin ter setup cost is higher. Under this conditions, b oth the lo cation up date and call deliv ery costs for the POFLA sc heme and the t w o-lev el p oin ter forw arding sc hemes increase. Since the p oin ter setup and tra v ersing costs for L p oin ter k eep unc hanged, the p erformance of the p eruser forw arding sc heme do es not c hange either. Ev en the costs of H p oin ter increase, the connection setup costs of the POFLA sc heme and the t w o-lev el p oin ter forw arding sc heme are still less than that for the p er-user forw arding sc heme (Fig. 2.10(b)). In Fig. 2.10(c), when the C M R is v ery lo w, the p er-user forw arding sc heme can p erform b etter; when the C M R is larger than one, the net costs for the three sc hemes are similar. Fig. 2.11 sho ws the p erformance for the three strategies when the p oin ter pro cessing cost P is increased to 0 : 1. With the increase of user p opulation and the users' mobilit y the pro cessing cost for p oin ter managemen t w ould increase to o. The pro cessing cost includes database transactions and ma y generate extra dela y with larger n um b er of op erational requests. As w e can observ e from Fig. 2.11, the costs for all three sc hemes increase. The POFLA sc heme will generate the least connection dela y and the total net p erformance is v ery close. In this c hapter, w e also compare the p erformance of our POFLA sc heme with that of the lo cal anc hor sc heme. The authors [22] ha v e suggested t w o v arian ts of the lo cal anc hor sc heme{the static one and the dynamic one. The static lo cal anc hor sc heme is easier to

PAGE 45

34 10 2 10 0 10 2 0.2 0.22 0.24 0.26 0.28 0.3 0.32 (a) Location update cost Calltomobility ratio POFLATwo_levelPeruser 10 2 10 0 10 2 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 (b) Call delivery cost Calltomobility ratio POFLATwo_levelPeruser 10 2 10 0 10 2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 (c) Net cost Calltomobility ratio POFLATwo_levelPeruser Figure 2.11: The relativ e costs for the three sc hemes with P = 0 : 1, G = 0 : 1 and = 1 : 5 10 2 10 0 10 2 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 (a) location update cost Calltomobility ratio POFLADLA 10 2 10 0 10 2 1 1.05 1.1 1.15 (b) call delivery cost Calltomobility ratio POFLADLA 10 2 10 0 10 2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 (c) net cost Calltomobility ratio POFLADLA Figure 2.12: The relativ e costs for the POFLA and DLA with P = 0 : 05, G = 0 : 1 and = 1 : 5

PAGE 46

35 10 1 10 0 10 1 10 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 The net cost Calltomobility ratio POFLAStatic local anchor Figure 2.13: The relativ e net costs for the POFLA and SLA with P = 0 : 05, G = 0 : 1 and = 1 : 5 implemen t. In this sc heme, a VLR serving a user is selected as the lo cal anc hor for that user and will not c hange un til the next call arriv es. In the dynamic lo cal anc hor sc heme, the user's curren t lo cal anc hor migh t c hange to the curren t serving one according to the user's exp ected future ev en ts. The dynamic lo cal anc hor sc heme is more dicult to implemen t than the static one; ho w ev er the results in [22] sho ws that the dynamic sc heme can guaran tee that the net cost is less than that of the basic IS-41 or GSM MAP strategy and the static sc heme migh t generate higher cost than the basic sc heme under some conditions. In fact, the lo cal anc hor sc heme is a sp ecial case of the POFLA sc heme. If w e let K 1 = 1, then the POFLA sc heme reduces to the dynamic lo cal anc hor sc heme. The p erformance comparisons of the dynamic lo cal anc hor with the POFLA sc heme are sho wn in Fig. 2.12. In order to mak e the comparison fair, the eectiv e p oin ter c hain length ( K 1 K 2 ) in the POFLA sc heme is same as the dynamic lo cal anc hor sc heme. In [22], the decision of the lo cal anc hor c hange is made based on the user's next ev en t, whic h is deriv ed according to the user's mobilit y pattern. In our analysis, w e assume the lo cal anc hor c hanges to mak e sure the net cost will not exceed the basic IS-41 sc heme cost. In Fig. 2.12, w e assume P = 0 : 05, G = 0 : 1, = 1 : 5 and m = F = 1. Based on these assumption, w e obtain the eectiv e p oin ter c hain length is four, so w e set K 1 = K 2 = 2. It can b e seen that in b oth the lo cation up date, call deliv ery

PAGE 47

36 10 2 10 0 10 2 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 (a) location update cost Calltomobility ratio g =0.01 g =1 g =100 10 2 10 0 10 2 1 1.05 1.1 1.15 1.2 1.25 (b) call delivery cost Calltomobility ratio g =0.01 g =1 g =100 10 2 10 0 10 2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 (c) net cost Calltomobility ratio g =0.01 g =1 g =100 Figure 2.14: The eect of v ariance of residence time ( r ) with P = 0 : 05, G = 0 : 1 and = 1 : 5 and the total net cost, the POFLA sc heme has b etter p erformance than the dynamic lo cal anc hor sc heme. In Fig. 2.13, w e also compare the total net costs of the POFLA sc heme with that of the static lo cal anc hor sc heme. In this gure, w e assume K 1 = K 2 = 3. W e can see that when the C M R is lo w, the static lo cal anc hor sc heme in v olv es v ery high trac load. 2.5.4 Sensitivit y to the RA Residence Time W e no w in v estigate the p erformance sensitivit y of POFLA sc heme to the v ariance of the RA residence time. W e assume that the RA residence time has a Gamma distribution. F or a Gamma distribution, the v ariance is V = 2 r i.e. a large r implies a small v ariance. Fig. 2.14 sho ws the eect of r on M 0 = M F 0 =F and C 0 =C resp ectiv ely In these gures, w e can observ e that the increase of the v ariance of RA residence time (smaller r ) causes the increase of M 0 = M and the decrease of F 0 =F ; the net eect to C F =C B is not signican t. The large v ariance implies that the n um b er of RA b oundaries the user crosses during the in ter-service time w ould v ary greatly If the user crosses man y RAs, a longer H p oin ter c hain will b e created. When the limit of H p oin ter K 2 is reac hed, the HLR will b e up dated, resulting in increase of M 0 On the other hand, if few er b oundaries are crossed, only shorter p oin ter c hains are set up, the p oin ter creation/tracing cost will b e sa v ed. The net eect is an increase in M 0 In Fig. 2.14(b), when the v ariance is high, if the crossed b oundary n um b er

PAGE 48

37 is small, the call deliv ery cost is reduced; If the n um b er is large, the p oin ter c hain could b e shortened b y H p oin ter or the up date to HLR. The net eect is a signican t impro v emen t in F 0 =F The net eect of the residence time v ariance on the total cost ratio C 0 =C is not signican t for all CMRs (see Fig. 2.14(c)). 2.6 Discussions In the t w o-lev el p oin ter forw arding and POFLA sc hemes, a p oin ter c hain can b e shortened b y the MAs the user visited. Th us, the up dates to the HLR is mitigated at the exp ense of the increase of the lo cal signaling trac. It is b enecial when the cost of comm unicating to HLR is relativ ely higher than the lo cal signaling cost. One adv an tage of the t w o new p oin ter forw arding strategies is that they can k eep the FIND p enalt y lo w while signican tly reducing the total system cost at the same time. As w e can see from the previous sections, the p er-user forw arding and lo cal anc hor sc hemes are the sp ecial cases of the t w o-lev el p oin ter forw arding and POFLA sc hemes. If w e adjust K 1 and K 2 prop erly the t w o new strategies reduce to the p er-user forw arding and lo cal anc hor sc hemes. There are some other w a ys to set up the p oin ter c hain. F or example, all the wireless comm unication service areas can b e divided in to Mobilit y Regions (MRs). A user will up date his/her lo cation to his/her MA in that region. Only when the user mo v es out of the region is a p oin ter set up from the old MA to the new one. F or the 3G wireless comm unication systems, 3GPP 23.119 sp ecication prop osed an approac h to limit the signaling trac b et w een the visited mobile system and the home mobile system [9, 51]. A new en tit y gateway lo c ation r e gister (GLR) is in tro duced b et w een the VLR/SGSN and the HLR. F rom the viewp oin t of the VLR/SGSN at the visited net w ork, the GLR is treated as the roaming user's HLR lo cated at the home net w ork. F rom the viewp oin t of the HLR at the home net w ork, the GLR acts as the VLR/SGSN at the visited net w ork. Indeed, in the 3G wireless systems, a new lev el of lo cation managemen t database is added. The users need to exc hange extra lo cal messages but reduce the long distance or in ternational messages exc hanged in eac h of the subsequen t registrations. The t w o-lev el p oin ter forw arding and POFLA strategy can b e implemen ted in the 3G systems in the follo wing manner: the GLRs can b e selected as the MAs. A realistic implemen tation of the new p oin ter forw arding sc hemes should also tak e in to accoun t the p ossibilit y that lo ops ma y form as the user visits sev eral RAs in succession. Th us, if a user revisits an RA and

PAGE 49

38 a p oin ter for that user is found in that VLR, then the old p oin ter can b e deleted to a v oid unnecessary op erations. This is called \implicit p oin ter compression" in [23]. 2.7 Conclusions In this c hapter, w e prop ose t w o new p oin ter forw arding lo cation managemen t sc hemes| t w o-lev el p oin ter forw arding and POFLA strategies, whic h in tend to reduce the cost of lo cation managemen t b y lo calizing or distributing the signaling trac and to o v ercome the HLR b ottlenec k problem while reducing the call setup (nding) dela y The p erformance analysis is carried out to sho w the adv an tages of the new prop osed sc hemes. Comparison studies with the p er-user p oin ter forw arding sc heme and the lo cal anc hor sc heme are also undertak en and sho w that the prop osed sc hemes outp erform either one of them. More imp ortan tly the prop osed sc hemes incorp orates more parameters to b e used to optimize the p erformance of the lo cation managemen t. Moreo v er, the prop osed sc hemes can b e easily tailored for the 3G wireless systems in whic h gatew a y lo cation register is in tro duced.

PAGE 50

CHAPTER 3 USER PR OFILE BASED STRA TEGY 3.1 In tro duction In the third generation wireless comm unication system, the high sp eed, real time and m ultimedia services will b e pro vided. The curren t RAs in the second generation systems will b e partitioned in to smaller areas in order to meet the QoS requiremen ts. It is ob vious that the small RA size will exacerbate the net w ork lo cation managemen t signaling burden signican tly to o. It can b e observ ed easily that most mobile users follo w some xed routes ev eryda y F or example, a p erson driv es to his oce ev ery morning along a road he usually c ho oses and sta ys in oce for most of the da y and then comes home after w ork; A mailman deliv ers mails along xed itinerary ev eryda y If the net w ork stores the mobile users' daily route information in the proles, then the lo cation up date signaling trac can b e minimized. Based on this idea, a Prole Based Sc heme (PBS) w as prop osed in [52, 53]. In this sc heme, a user's daily routine information is stored in the prole. If the user follo ws his or her itinerary w ell then no lo cation up date message is sen t so that the up date trac is reduced. When a call arriv es for that user, all the RAs the user could b e in will b e paged. The paging can b e implemen ted in all the RAs at the same time or op erated one b y one follo wing a descending probabilit y order. The latter option is suggested in [53]. It is straigh tforw ard that the lo cation up date cost is reduced at the exp ense of increasing the total paging costs. A user who follo ws some daily routine ma y deviate from his or her usual course b ecause of road trac, w eather or other reasons. If this happ ens, the mobile terminal is required to rep ort to the new VLR ev ery time just lik e in the IS-41 or GSM MAP sc heme. So the mobile terminal needs to store the RA list information in memory to o and to b e up dated or adjusted b y the net w ork p erio dically In this c hapter, w e prop ose a new sc heme whic h is called user Mobilit y P attern Based Lo cation Sc heme (MPBS) to impro v e the system lo cation managemen t p erformance. In the 39

PAGE 51

40 new sc heme, the User Mobility Pattern (UMP) is considered and retriev ed for lo cation up date and call deliv ery pro cedures. The sim ulation results sho w that the MPBS generates less signaling trac than the IS-41 and PBS sc hemes and the paging dela y in the MPBS is also less than that in PBS. In the next generation wireless comm unication systems, the service pro viders w an t to pro vide users the user-orien ted services, namely the system resources are allo cated according to the user's b eha viors [54{ 56]. This will require the UMP to b e considered carefully F or example, in the MPBS sc heme, the system can predict a user's future lo cation so that resource can b e assigned in adv ance if the user is engaging in some imp ortan t applications. These pro cedures can impro v e the QoS greatly In this c hapter, w e giv e the detail description ab out the MPBS sc heme and compare the p erformance with IS-41 and PBS sc heme b y sim ulations. In [53], the author did not consider the Call-to-Mobilit y Ratio ( C M R ) eect on PBS p erformance. In our w orks, w e also sho w that the user CMR is v ery critical to the PBS and trac sa ving can only b e ac hiev ed under some limited CMR range. This c hapter is organized as follo ws. In section 3.2, w e describ e the system arc hitecture to implemen t our sc heme and in tro duce the VMN concept; Section 3.3 giv es the details of the PBS and MPBS sc hemes; W e analyze the costs and presen t the sim ulation congurations for the new sc heme in section 3.4; Section 3.5 presen ts the sim ulation results and compares the sc heme p erformances under v arious conditions; The conclusion is dra wn in section 3.6. 3.2 System Description F or the next generation wireless m ultimedia net w orks, dieren t kinds of users ha v e dieren t service requiremen ts. Dieren t services and QoS can b e assigned to users based on their past call or mobilit y history and their willingness of pa ying for a higher QoS. In order to attract users, the net w ork needs to store some user information in his/her prole and retriev e it to nd out what kind of service can meet the user's requiremen t. If the system stores the necessary information in the prole, some system resource can b e reserv ed in adv ance in order to pro vide the user real time service. F urthermore, if the user mobilit y pattern is kno wn, some sp ecic information can b e prepared for a sp ecic user in some sp ecic area the user will b e in so on. If the ab o v e w orks, lik e information collection and dissemination, are incorp orated in to the curren t distributed wireless comm unication database system, the

PAGE 52

41 signaling and database retriev al trac will o v erwhelm the system pro cess capacit y The dela y resulted from the hea vy trac load will degrade the whole system p erformance. In order to solv e the problem, w e prop ose the Virtual Managemen t Net w ork (VMN) concept. VMN is an o v erla y managemen t net w ork on top of the existing wireless mobile net w ork, whic h handles the in telligen t features. In our sc heme, a new function en tit y Mobilit y Agen t (MA), is prop osed. If w e assume the signaling net w ork for a wireless mobile net w ork is a Common Channel Signaling (CCS) net w ork, lik e Signaling System No.7 (SS7), the MAs can b e hardw are-based or soft w are-based and connect to b oth VLRs and HLRs. A t the same time, the MAs are connected to eac h other. Eac h MA is in c harge of a selected set of lo cation areas and can b e dynamically congured according to the up-to-date trac burden. In SS7, the MAs can b e placed together or separately at the same lev el as LSTP or RSTP If b oth LSTP and RSTP are equipp ed with MAs, w e can sa y that the VMN is congured hierarc hically The VMN can also connect to O A&M cen ter. 3.3 PBS and MPBS Sc hemes In this section, w e describ e the MPBS pro cedures in details; w e also in tro duce the PBS pro cedures generally and the details ab out the sc heme can b e found in [53]. 3.3.1 Prole Based Lo cation Sc heme The prole based lo cation sc heme (PBS) w as prop osed b y WINLAB, Rutgers Univ ersit y In this sc heme, the system main tains records of eac h user's most lik ely itinerary list. It is assumed that the user lo cation distribution probabilit y is kno wn in adv ance. It can b e pro vided b y the mobile terminal or estimated b y the system according to the user's past calling history The lo cation list is stored in the switc h that will conduct the searc h for the user and the information required to up date the list is stored within the mobile switc hing cen ter's billing records. The user's itinerary can b e dened as follo ws. If A i is one of the lo cation areas in the record list, the user's most lik ely itinerary can b e dened as f A i g ki =1 where k is the elemen t n um b er in the set. The probabilit y of a user b eing in a lo cation area A i is i The system main tains a list of ( A i ; i ) pairs for some time in terv al T The probabilit y of a user b eing out of f A i g ki =1 is giv en as = 1 k X i =1 i : (3.1)

PAGE 53

42 If > 0, there is alw a ys a probabilit y of not nding the user. If the user follo ws his/her daily itinerary strictly namely the user k eeps roaming in f A i g ki =1 no registration is needed. When the user is out of the list, the mobile terminal is required to man ually register to the system. So the terminal m ust k eep a cop y of the list. The list can b e sen t to the terminal b y system and up dated only when some A i is added or deleted. When a call arriv es for the user, the lo cation areas in the list are paged in descending order of i un til the user is found. Under this strategy the database will kno w a user's exact lo cation when it is out of f A i g ki =1 Indeed, when the user is out of the list, the PBS sc heme p erforms the same as IS-41 or GSM MAP According to pro cedures, The PBS can reduce the lo cation up date cost eectiv ely at the exp ense of increasing paging cost or paging dela y In [53], the authors only studied the paging dela y and the radio link cost under dieren t list length. The paging dela y w as deriv ed, giv en three kno wn probabilit y distributions, in the term of the exp ected lo cation area n um b ers needed to b e paged b efore the user can b e found. In fact, it is in tuitiv e that the total cost of PBS sc heme has tigh t relationship with the user's C M R With small C M R whic h means the user has relativ e higher mo ving rate than call arriv al rate, the PBS can reduce most up date cost and the nal result is go o d. While, for the users with high C M R the paging cost will b e dominan t and the total cost of PBS ma y exceed that of the basic IS-41 or GSM MAP sc heme. 3.3.2 Mobilit y P attern Based Sc heme In order to impro v e the PBS sc heme p erformance, w e prop ose a new mobilit y managemen t strategy|mobilit y pattern based sc heme (MPBS) in this c hapter. The MPBS strategy can reduce the user up date cost and try to limit the paging cost at the same time. Comparing with PBS, only t w o more elemen ts, the time a user en tering A i and the residence time in A i are added in the user prole. The MA will k eep a list of 4-tuple ( A i ; t i ; T i ; i ) for eac h user. W e assume the cardinalit y is k In the list, the tuples is not ordered according to i but to t i F or example, if a user's prole includes all the lo cation en tries he/she ma y visit in 24 hours, the list is sorted b y the time the user visits ev ery lo cation area. So for i 6 = j A i and A j ma y b e same. In MPBS sc heme, w e dene the user out-of-pattern probabilit y whic h is giv en as = 1 k X i =1 i : (3.2)

PAGE 54

43 Restaurant A1A2A3A4A5t1t2t3t4t5T1T2T3T4t6T5 Office Home Figure 3.1: Mobilit y pattern based sc heme pro cedures If > 0, there is probabilit y that the user mo v es out of the lo cation areas in the list. W e sa y the user is out-of-patter if this happ ens; otherwise, w e sa y the user is in-the-pattern. Fig. 3.1 sho ws an example of the MPBS. If a user liv es in lo cation area A 1 and w orks in his oce building in A 4 Before the user reac hes his oce, he will pass through A 2 and A 3 ; The user also lik es to ha v e his lunc h break in a restauran t lo cated in A 5 If the user follo ws the itinerary ev eryda y and the information is stored in his prole, the net w ork can lo cate the user based on the prole and curren t system time. T o mak e the sc heme clear, w e need to dene the user b eha viors more precisely In the MPBS sc heme, when users follo w the UMP the lo cation up date trac can b e reduced. There are t w o kinds of patterns for users to follo w in the MPBS sc heme. When user en ters A i at time t i and the residence time in A i is T i w e sa y the user follo ws time-sequence pattern. If the user en ters and exits lo cation areas follo wing the A i order in the prole only w e sa y the user follo ws the sequence pattern. It is ob vious that a user follo wing the time-sequence pattern m ust follo w the sequence pattern as w ell, but a user follo wing the sequence pattern ma y not follo w the time-sequence pattern. Ho w ev er, users ma y deviate from their daily routines b ecause of w eather or road trac situations. So w e need to nd out ho w close a user follo ws his/her mobilit y pattern. The mobile terminals for the next generation systems should incorp orate more in telligen t functions. When a user roams in the net w ork service areas, w e assume that the mobile terminal can record the lo cation area ID, lo cation area

PAGE 55

44 en trance and exit time. W e dene the user's actual path information as the User A ctual Path (UAP). The UAP can b e used to up date the UMP p erio dically As the time elapses or user crosses lo cation area b oundary the mobile terminal can tell whether the user is follo wing an y pattern or not b y comparing the UAP with UMP The UAP is in the same format as UMP If w e do not consider the time information, the UAP can b e expressed as f B i g mi =1 where m is the length of UAP Without the consideration of time, w e can compare the similarit y b et w een UAP and UMP b y edited distance [57]. W e assume the regular mo v emen t of a mobile user can b e mo delled as an edited UAP b y allo wing the follo wing legal op erations: Inserting a lo cation area L at p osition i of the UMP giv es UAP: A 1 ; A 2 ; ; A i 1 ; L; A i ; A i +1 ; ; A k Deleting the lo cation area A i of the UPM giv es UAP: A 1 ; A 2 ; ; A i 1 ; A i +1 ; ; A k Changing a lo cation area A i of the UMP to another lo cation area L giv es UAP: A 1 ; A 2 ; ; A i 1 ; L; A i +1 ; ; A k As a result, the edited distance b et w een a UMP and a UAP b ecomes the sum of the w eigh ts of the editing op erations. If the edited distance is less than a threshold, w e sa y the user follo ws the sequence pattern, indicating the general mo ving in ten tion of the user. F or large systems with complex net w ork top ologies, the calculation of the spatial w eigh ts can b e quite in v olv ed. Ho w to assign and calculate the spatial w eigh t exactly is out of the scop e of our discussion. F or simplicit y and without loss of generalit y w e can dene the w eigh t as follo ws: The cost of inserting W I = 8><>: 1 L is the adjacen t lo cation area of A i 1 otherwise The cost of deleting W D = 8><>: 0 A 1 , A i 1 ha v e already b een deleted 1 otherwise The cost of c hanging W C = 8><>: 1 L is the adjacen t lo cation area of A i 1 otherwise

PAGE 56

45 Based on the ab o v e notations, w e can dene user b eha viors more precisely When a user en ters a lo cation area A i the user is said to follo w the time-sequence pattern if and only if the follo wing requiremen ts are met: (1) A i 2 f A i g ; (2) j t i;actual t i j < T and (3) t t i T i < T where t i;actual is the actual time the user en ters A i t is the curren t system time and T is the time pattern threshold. The rst condition constrains the user in the prole, the second and third conditions limit the user to en ter and exit lo cation area A i within some time threshold. If w e assume a UAP is f B i g mi =1 the edited distance b et w een UAP and UMP is d ( A 1 ; A 2 ; ; A m ; B 1 ; B 2 ; ; B m ) and the edited distance threshold is D w e can sa y the user follo ws the sequence pattern if: (1) A i 2 f A i g ; (2) j t i;actual t i j > T or t t i T i > T and (3) d ( A 1 ; A 2 ; ; A m ; B 1 ; B 2 ; ; B m ) =m < D In MPBS sc heme, users can b e in one of the four states when en ter a lo cation area A : State 1: If a user follo ws the time-sequence pattern, w e dene the user in state 1. State 2: if a user follo ws the sequence pattern, w e dene the user in state 2. State 3: if a user do es not follo w an y of the ab o v e t w o patterns but A 2 f A i g w e dene the user in state 3. State 4: if A = 2 f A i g w e dene the user in state 4. The reason w e dene four states is that the system can in v ok e dieren t paging mec hanisms for users in dieren t states. A user needs to register to the system when he/she switc hes states. In the MPBS sc heme, w e assume the user state information is included in the up date message, so the system can kno w the user curren t state ev ery time the mobile terminal sends up date message. When a user is in state 1, 2 or 3, no registration or state message needs to b e sen t if the user k eeps the state unc hanged. If the user k eeps in state 4, the terminal will up date its lo cation to the system ev ery time the user en ters a new lo cation area. Then, only in state 4, the user needs to up date the lo cation ev ery time he/she crosses an RA b oundary In state 1, state 2 and state 3, the user needs to send up date message only when the states switc h. Then if w e can collect the user daily routine information w ell so that the user has large probabilit y in state 1,2 or 3, the up date cost can b e reduced. In the MPBS sc heme, when a call arriv es, the system can adopt dieren t paging strategies based on the user's curren t state. If the user is in state 1, the system can decide whic h lo cation area the user is in according to the curren t system time and page it. If the curren t time is t and the user

PAGE 57

46 0 1 2 3 4 5 6 7 8 x 10 4 0 10 20 30 40 50 60 Time (second)Update unmber in 24 hours MPBSIS41PBS Figure 3.2: The up date n um b ers for three sc hemes in 24 hours with user residence time 30 min utes is in state 1, the system can retriev e the tuple matc hing t i t t i + T i then the user can b e found in A i There is the probabilit y that the user just mo v es out of the paged area and en ters the next one. If the users do es not c hange state, only the next lo cation area needs to b e paged. If the user is in state 2, the system kno ws the lo cation areas the user is not in according to the last lo cation up date. Since the user follo ws the sequence pattern, the system kno ws the user m ust b e in one of the lo cation areas after the last up dated one in the prole. Because there is no time information in state 2, all the lo cation areas the user could b e in will b e paged in the descending order of the probabilities i If the user is in state 3, the system kno ws that the user is in f A i g and all the lo cation areas in f A i g will b e paged according to the descending order of i un til the user is found whic h is just lik e the PBS sc heme. If the user is state 4, the system kno ws the user's exact lo cation area and pages. In fact, when the user is in the state 4, the MPBS sc heme is exactly the same as IS-41/GSM MAP sc heme. As w e can see from the three sc hemes, the IS-41 will generate the most up date messages, the PBS sc heme will generate the least ones and the MPBS is in the middle of them. Fig. 3.2 sho ws this clearly Although the MPBS generates more up date messages than the PBS, it reduces the paging cost dramatically than the PBS and ac hiev es total cost reduction.

PAGE 58

47 3.4 Cost Ev aluation and Sim ulations In this section, w e try to estimate the total signaling costs for the three lo cation managemen t sc hemes. Since the user b eha viors and man y other random factors will aect the results signican tly it is hard to get the close form of the cost functions [58{ 60]. So, in this c hapter, w e just giv e the general estimation for eac h sc heme signaling cost. The sc heme p erformance comparison is carried out b y sim ulations. W e dene C u the cost for lo cation registration and C p the cost for paging one lo cation area. In our analysis, w e do not consider the call deliv ery pro cessing cost b ecause it is same for all the three sc hemes. F or IS-41/GSM MAP sc heme, the total cost b et w een t w o consecutiv e calls can b e giv en b elo w [45] C GS M = C u + C p ; (3.3) where is the C M R In the follo wing analysis, w e assume there are k lo cation areas in the user's prole. F or the PBS sc heme, w e dene E ( k ) the a v erage lo cation area n um b er that has to b e paged b efore the user is found. So it is straigh tforw ard that the total cost for PBS sc heme is C P B S = 1 C u + (1 ) E ( K ) C p + C p = 1 C u + [ (1 E ( k )) + E ( k )] C p ; (3.4) where 1 is the probabilit y that the user mo v es in and out of the prole b et w een t w o consecutiv e call arriv als in the PBS sc heme. The analysis of total cost for MPBS is more complicated. If w e dene C p i the paging cost for users in state i and i the probabilit y that the user is in state i when a call arriv es, resp ectiv ely the total cost of MPBS can b e expressed: C M P B S = 2 C u + 4 X i i C p i ; (3.5) where 2 is the probabilit y of the user's mo v emen ts that need to up date to the HLR. It is ob vious that P 4i i = 1. W e need to deriv e C p i to sp ecify the total cost of MPBS sc heme. In the MPBS, when user is in state 1, if a call arriv es, the lo cation area the user is curren tly in will b e paged

PAGE 59

48 according to the prole. Ho w ev er, there is probabilit y that the user just mo v es in to the next lo cation area when the curren t lo cation area is paged. In order to mak e sure the user can b e found, the next lo cation area needs to b e paged if no resp onse is receiv ed in predened time in curren t area. W e assume that t i;actual t i follo ws Gaussian distribution with zero mean and v ariance If j t i;actual t i j > T w e sa y the user is not in state 1 an ymore. Then the probabilit y that the user just mo v es to the next lo cation area when a page message arriv es is 0 : 5 Q ( T ). Since the paging expiration time is v ery short, w e ignore the probabilit y that the user crosses more than one lo cation areas when the paging message arriv es. Based on the ab o v e results, w e can obtain C p 1 = C p + [0 : 5 Q ( T )] C p = [1 : 5 Q ( T )] C p : (3.6) If a user is in state 2 when a call arriv es, only the follo wing lo cation areas need to b e paged b ecause it is kno wn that the user is not in the rst m ( m < k ) lo cation areas. So the distribution probabilit y in the follo wing lo cation areas is conditional. If w e assume there are m lo cation areas has b een crossed b efore a call arriv es, and let X m = 1 + 2 + + m and k 0 = k m w e can get the conditional probabilit y distribution for the next k 0 lo cation areas as 0 1 = m +1 1 X m 0 2 = m +2 1 X m 0 k 0 = k 1 X m resp ectiv ely If the a v erage paged n um b er is E ( k 0 ), the C p 2 can b e written as C p 2 = E ( k 0 ) C p : (3.7) In the real situation, it is ma y b e dicult to get the conditional probabilities. But in MPBS sc heme, there is no necessit y for the system to compute them b ecause the system only needs to page the follo wing lo cation areas in the descending order of m +1 m +2 , k Divided b y a same p ositiv e v alue do es not aect their order. The cost of C p 3 is just lik e PBS sc heme and the cost of C p 4 is the same as IS-41/GSM MAP sc heme. So w e ha v e C p 3 = E ( k ) C p ; (3.8) and C p 4 = C p : (3.9)

PAGE 60

49 RA15 RA11 RA10 RA9 RA8 RA7 RA6 RA5 RA4 RA3 RA2 RA1 RA14 HLR RA0 RA12 RA13 Figure 3.3: Sim ulation net w ork arc hitecture So, w e can rewrite (3.5) as C M P B S = 2 C u + [ 1 (1 : 5 Q ( T )) + 2 E ( k 0 ) + 3 E ( k ) + 4 ] C p : (3.10) In our sim ulations, w e treat the state up date message the same as the registration message and normalize the up date cost C u = 1. The paging cost is usually less than up date cost, so w e assume the paging cost for one lo cation area is 0 < C p C u W e are in terested in ho w the C M R the paging cost C p the user distribution probabilit y i and the user out-of-pattern probabilit y aect the p erformance of the PBS and MPBS sc hemes. In this part, w e assume there are three represen tativ e prole distribution [53]: uniform, linear and exp onen tial. Denote the conditional probabilit y of a user b eing in the ith lo cation area in the list as i = i = (1 ) or i = i = (1 ). The denitions for the three kinds of distribution are giv en as b elo w. Uniform Distribution: when 1 = 2 = = k = 1 =k the prole is said to b e uniformly distributed. Linear Distribution: when i = 2( k +1 i ) k ( k +1) for i 2 f 1 ; 2 ; ; k g the prole is said to b e linearly distributed. Exp onen tial Distribution: when i = e bi (1 e b ) e b e b ( k +1) for i 2 f i; 2 ; ; k g the prole is said to b e exp onen tially distributed, where b is a constan t. The sim ulation net w ork arc hitecture is sho wn in Fig. 3.3. The net w ork consists of 16 RAs. Eac h RA has 4 neigh b oring RAs. The links b et w een RAs are virtual links, whic h imply the user can roam bidirectionally b et w een the connected RAs. The links b et w een RAs

PAGE 61

50 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 The user CalltoMobility ratioThe update cost ratio of MPBS to GSM Figure 3.4: The lo cation up date cost ratio of MPBS sc heme to IS-41/GSM MAP and HLR are signaling links [61]. When the user en ters an RA, dep ending on the sc hemes curren tly adopted, an up date message ma y b e generated and sen t to HLR. The user resides in an RA for some exp onen tially distributed time with predened mean, then mo v es in to the next RA [62]. The calls for that user generated in HLR forms a P oisson pro cess. F or simplicit y and clarit y w e em b ed the VMN functions in to the HLR no de. The HLR no de will record all the up date and paging costs for dieren t sc hemes at the end of the sim ulations. In the sim ulations, w e assume the up date cost for ev ery lo cation area is same. The sim ulations are ev en t driv en. In the initiation, a user is generated randomly in one lo cation area and set in state 1. The state 4 is dieren t from the other states. In state 4, the user in fact do es not follo w an y pattern. So when the user is in state 1, 2 or 3, w e sa y the user is in-the-pattern, and the user is out-of-pattern when he/she is in state 4. W e also assume 90 p ercen t of the time, the user is in-the-pattern. The conditional probabilit y of the user in state 1, state 2 and state 3 are 0.8, 0.15 and 0.05, resp ectiv ely All the sim ulations collect the user trac king information for 24 hours. In order to study the out-of-pattern probabilit y aection on the sc heme p erformance, the will c hange during the sim ulations. F or simplicit y but no loss of generalit y w e assume i and t i are in the same order in the prole and A i 6 = A j for i 6 = j

PAGE 62

51 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0.26 0.28 0.3 0.32 0.34 0.36 0.38 0.4 0.42 0.44 The user CalltoMobility ratioThe MPBS to PBS user locating time ratio Paging cost=0.1Paging cost=0.5 Figure 3.5: The comparison of the lo cating time for MPBS and PBS 3.5 Numerical Results and Comparison As w e men tioned b efore, the MPBS can reduce the lo cation up date signaling trac. Fig. 3.4 sho ws the relativ e lo cation up date cost of the MPBS to IS-41 or GSM MAP sc heme. In this sim ulation, the out-of-pattern probabilit y = 0 : 1 and the probabilities of user in three dieren t states are set as ab o v e. In Fig. 3.4, the up date cost of MPBS sc heme is less than half of the up date cost in the IS-41/GSM MAP sc heme, and the C M R do es not aect the p erformance dramatically The up date cost is usually larger than the paging cost. That is ho w the PBS and MPBS can ac hiev e total sa ving b y reducing the up date cost at the exp ense of increasing the total paging cost. Ho w ev er the MPBS can reduce the total cost without increasing the paging cost to o m uc h. In b oth PBS and MPBS sc hemes, the system usually pages more than one lo cation areas trying to nd out the user's exact lo cation. In other w ords, the MPBS and PBS will in tro duce some dela y during call deliv ery pro cedures. In Fig. 3.5, w e can see that the paging dela y generated b y MPBS is m uc h less than PBS. The reasons are follo ws. In PBS sc heme, all the lo cation areas in the list are needed to b e paged. Usually the lo cation areas are paged sequen tially according to the user distribution probabilit y So the dela y is the time elapsed from the paging messages is sen t to the rst lo cation area to the user resp onses. The dela y can b e dieren t with dieren t probabilit y distributions. In Fig. 3.5, w e assume the user prole distribution is uniform. In the MPBS

PAGE 63

52 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 The user CalltoMobility ratioThe ratios of total cost to GSM with uniform distribution MPBS with finding cost=0.1PBS with finding cost=0.1MPBS with finding cost=0.5PBS with finding cost=0.5 Figure 3.6: The total costs of MPBS and PBS to IS-41/GSM MAP with uniform distribution sc heme, when the user is in-the-pattern, only when the user is in state 3, need all lo cation areas to b e sequen tially paged. When user is in state 1, only one lo cation area is paged in most of time; when in state 2, only part of the f A i g need to b e paged. The paging cost could b e considered as the paging dela y for one lo cation area. In our sim ulations, w e assume the paging cost for ev ery lo cation area is equal. As w e can see, when the C M R is lo w, the paging dela y for MPBS sc heme is 70% less than PBS sc heme. The reason is that when the C M R is lo w, the user has relativ e high mo v emen t probabilit y the PBS sc heme will page more lo cation areas trying to nd out the user; but the MPBS sc heme is not aected b y this factor. When the C M R is large, the user will sta y in a lo cation area for a relativ e long time, the PBS sc heme can nd out the user with less lo cation area paging. Ho w ev er, the MPBS total paging cost is still 60% less than PBS sc heme. Although the MPBS has less paging dela y than PBS, w e need to examine the total costs for the t w o sc hemes and try to see whether they can ac hiev e b etter p erformance than con v en tional IS-41/GSM MAP sc heme or not. In Fig. 3.6, 3.7 and 3.8, w e plot b oth MPBS and PBS to IS-41/GSM MAP total cost ratios with three dieren t probabilit y distributions and dieren t paging costs. W e can see from these gures, the costs for PBS sc heme increase v ery quic kly with the increase of paging cost. When the paging cost is 0.5, the PBS sc heme can ha v e sa ving only when the CMR is v ery lo w. In that situation, the user will mak e

PAGE 64

53 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 The user CalltoMobility ratioThe ratios of total cost to GSM with linear distribution MPBS with paging cost=0.1PBS with paging cost=0.1MPBS with paging cost=0.5PBS with paging cost=0.5 Figure 3.7: The total costs of MPBS and PBS to IS-41/GSM MAP with linear distribution 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 The user CalltoMobility ratioThe ratios of total cost to GSM with exponential distribution MPBS with paging cost=0.1PBS with paging cost=0.1MPBS with paging cost=0.5PBS with paging cost=0.5 Figure 3.8: The total costs of MPBS and PBS to IS-41/GSM MAP with exp onen tial distribution

PAGE 65

54 a lot of up dates to the system and the up date cost is dominan t. With the increase of C M R the paging cost for PBS sc heme pla ys a more imp ortan t role, then the total cost increases fast. The MPBS total cost increases m uc h slo wly with the paging cost. This is the adv an tage of MPBS. In the wireless m ultimedia net w orks, the data service will consume a lot of bandwidth, whic h mak es the paging cost higher. The MPBS total cost increases m uc h slo w er than PBS do es with the increase of paging cost. Another adv an tage of MPBS is that the cost curv es are ratter than PBS in a wide range of CMR when the paging cost is large. That means the MPBS sc heme is applicable to dieren t classes of users with dieren t mobilit y pattern. In Fig. 3.8, the total cost of PBS sc heme is a little less than MPBS sc heme when the paging cost is small. The reason is that, for exp onen tial distribution, the paged lo cation areas is less than other distributions. When the paging cost is small, the total cost ma y b e small. But ev en with exp onen tial distribution, the PBS total cost increase m uc h faster than MPBS sc heme do es with the paging cost. In the ab o v e situations, w e assume 90 p ercen t of the time, the user is in-the-pattern. It is in tuitiv e that the total costs of b oth MPBS and PBS sc hemes ha v e imp ortan t relationship with the out-of-pattern probabilit y lik e or Fig. 3.9 and Fig. 3.10 sho w ho w the user out-of-pattern probabilit y aects the system p erformance for b oth MPBS and PBS sc hemes with dieren t probabilit y distributions. In b oth gures, w e assume C M R = 1. The paging cost is 0 : 1 in Fig. 3.9 and 0.5 in Fig. 3.10. In Fig. 3.9, b oth the PBS and MPBS sc hemes total costs increase as the user out-of-pattern probabilit y increases. The cost of MPBS is less than PBS. When the paging cost is large, as in Fig. 3.10, the total costs for PBS are larger than 1 for an y one of the three distributions. This also pro v es the conclusion w e made b efore, the PBS sc heme can impro v e the system p erformance only in v ery limited conditions. In Fig. 3.10, the total costs of MPBS are still less than one ev en the user has large probabilit y to get out of the pattern. In the t w o gures, all the cost ratios are equal to one when the user out-of-pattern probabilit y is one. The reason is ob vious. When the user out-of-pattern is one, the user will not en ter an y lo cation area in the prole list and needs to register to the system ev ery time he/she mo v es. The system mobilit y managemen t sc heme is the same as IS-41/GSM sc heme indeed.

PAGE 66

55 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 The user outofpatternprobabilityThe ratios of total cost to GSM with paging cost=0.1 MPBS with uniform dist.MPBS with linear dist.MPBS with exponentional dist.PBS with uniform dist.PBS with linear dist.PBS with exponentioanl dist. Figure 3.9: The eects of user out-of-pattern probabilit y on MPBS and PBS with pag ing cost = 0 : 1 ; C M R = 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 The user outofpatternprobabilityThe ratios of total cost to GSM with paging cost=0.5 MPBS with uniform dist.MPBS with linear dist.MPBS with exponentional dist.PBS with uniform dist.PBS with linear dist.PBS with exponentioanl dist. Figure 3.10: The eects of user out-of-pattern probabilit y on MPBS and PBS with pag ing cost = 0 : 5 ; C M R = 1

PAGE 67

56 3.6 Conclusions F or the next generation wireless m ultimedia comm unication systems, the radio sp ectrum is the scarcest resource. In order to exploit the radio resource and pro vide users services more ecien tly the lo cation areas b ecome smaller. The result is that the user lo cation up date message will consume a lot of bandwidth. This situation b ecomes w orse with the increase of user n um b er. Man y researc h w orks ha v e b een carried out to optimize the system p erformance in user mobilit y managemen t area, lik e the PBS sc heme. In PBS, no up date message needs to b e sen t as long as the user is in the prole list. Only when the user gets out of the prole, need registration messages b e sen t to the system. When a call arriv es for the user, the lo cation areas in the prole are paged according to the descending order of distribution probabilit y un til the user is found. In the PBS sc heme, the user lo cation up date message cost is sa v ed at the exp ense of increasing the paging cost. In this c hapter, w e studied the PBS sc heme p erformance based on dieren t probabilit y distributions, paging costs and C M R s The aection of the user's out-of-pattern probabilit y is also in v estigated. The results sho w that PBS sc heme only w orks w ell for v ery small C M R and the total cost increases quic kly with the paging cost. W e also prop osed a new lo cation strategy|MPBS in this c hapter. In MPBS sc heme, the user mobilit y time pattern is recorded in the user's prole to o. When user is in-the-pattern, there are three states the user could b e in. The user up dates his/her lo cation only when the states or pattern c hange. The sim ulation results suggest that, although the MPBS sc heme generates more up date messages than PBS do es, the total cost of MPBS is usually signican tly less than PBS and the MPBS sc heme is not v ery sensitiv e to the increase of paging cost either. This is v ery imp ortan t b ecause the paging op eration will b ecome costly in the m ultimedia net w orks where wide bandwidth services will b e pro vided. One of the most imp ortan t QoS factors is the connection setup time. In our system arc hitecture, the setup time is em b o died b y the paging dela y Our sim ulation results sho w that the paging dela y of MPBS sc heme is o v er 50% less than PBS sc heme. Our results also sho w that MPBS sc heme can w ork w ell for user with dieren t C M R In this c hapter, w e also prop ose the system arc hitecture to implemen t MPBS sc heme and a new no de item, Mobilit y Agen t (MA), is in tro duced. The MAs form a new supp ort net w ork for the wireless comm unication system and connect to the system databases and op eration-con trol cen ter.

PAGE 68

57 The MAs can collect and pro cess the users' history information in their c harge areas and pro vide the user more sp ecic services.

PAGE 69

CHAPTER 4 MOBILITY MANA GEMENT IN MOBILE IP NETW ORK 4.1 In tro duction The curren t fast increasing demand for wireless access to in ternet applications is fueled b y the remark able success of wireless comm unication net w orks and the explosiv e gro wth of the In ternet. The future generation wireless net w orks target to pro vide users with high-sp eed in ternet access and more sophisticated services b esides v oice comm unication services [63]. The user equipmen t, suc h as wireless laptops, cellular phones and palm pilots, mak e it p ossible for mobile users to access the in ternet applications that predominan tly based on IP tec hnology [64, 65]. The p opularit y of the In ternet pro vides strong incen tiv es to service pro viders to supp ort seamless user mobilit y Ho w ev er, man y telecomm unications systems suc h as rst and second generation wireless cellular systems w ere designed mainly for v oice services, the in tegration with data net w orks b ecomes the ma jor push for third generation and future generation wireless systems. Mobile IP is the mobilit y-enabling proto col dev elop ed b y the In ternet Engineering T ask F orce (IETF) to supp ort global mobilit y in IP net w orks [11, 16]. This standard has b ecome the solution to solv e the user mobilit y in almost all wireless mobile systems. The IP proto col has b een designed for wired net w orks. There are t w o ma jor functions for the terminal IP address in the In ternet. An IP address is used to iden tify a particular end system in the whole net w ork and is also used to nd a route b et w een the endp oin ts. In IP net w orks, the pac k ets deliv ered to a particular end system are routed based on the destination IP address b y the in termediate routers. Based on this observ ation, w e can conclude that a mobile terminal needs to ha v e a stable IP address in order to b e stably iden tiable to other net w ork no des and also needs a temp orary IP address for the routing purp ose. The Mobile IP proto col extends IP b y allo wing a mobile no de to eectiv ely utilize t w o IP addresses, one for iden tication and the other for routing. 58

PAGE 70

59 Mobile IP enables mobile terminals to main tain all ongoing comm unications with the In ternet while mo ving from one subnet to another. In the Mobile IP proto col, mobile terminals that can c hange their p oin ts of attac hmen t in dieren t subnets are called Mobile Hosts (MHs). An MH has a p ermanen t address (home address) registered in its home net w ork and this IP address remains unc hanged when the user mo v es from subnet to subnet. This address is used for iden tication and routing purp ose, whic h is stored in a Home Agen t (HA). A HA is a router in a mobile no de's home net w ork, whic h can in tercept and tunnel the pac k ets for the mobile no de and also main tains the curren t lo cation information for the mobile no de. If an MH roams to a subnet w ork other than the home net w ork, this subnet w ork is a foreign net w ork for that user. In the curren t Mobile IP proto col, the MH can obtain a new IP address from a router in the visited net w ork. The router in the visited net w orks whic h assigns MH the IP address is the MH's foreign agen t (F A) and the new address is the MH's care-of address (CoA) used for pac k et routing purp ose. The CoA for the MH will c hange from subnet to subnet. In order to main tain con tin uous services while the user is on the mo v e, Mobile IP requires the MHs to up date their lo cations to the HAs whenev er they mo v e to dieren t subnets so that the HAs can in tercept the pac k ets deliv ered to them and tunnel the pac k ets to the user's curren t p oin ts of attac hmen t. Th us, the Mobile IP can pro vide con tin uous in ternet access services for mobile users and do es pro vide a simple and scalable solution to user mobilit y Ho w ev er, Mobile IP is not a go o d solution for users with high mobilit y Its mec hanism requires ev ery MH to up date its new CoA to the HA ev ery time the MH mo v es from one subnet to another, ev en though the MH dose not comm unicate with others while mo ving. As sho wn in Fig. 4.1, the lo cation up date cost in Mobile IP can b e excessiv e, esp ecially for the mobile users with relativ ely high mobilit y and long distance from their HAs. This problem b ecomes w orse with the increase of the mobile user n um b er [66]. Moreo v er, if a user is far a w a y from his/her home agen t or the HA pro cessing capabilit y is o v erwhelmed b y the h uge v olume of up date messages, the signaling dela y for the lo cation up date could b e v ery long, whic h will result in the loss of a large amoun t of in-righ t pac k ets and degrade the Qualit y of Service (QoS) [67].

PAGE 71

60 FA1FA2FA3FA4FA5FA6Location registration Packet delivery MH movement HA Home Agent FA Foreigen Agent MH Mobile Host HA MH MH MH INTERNET Figure 4.1: The MIP lo cation registration and pac k et routing In this c hapter, w e prop ose a dynamic hierarc hical mobilit y managemen t sc heme (DHMIP) for Mobile IP net w orks. In our sc heme, the lo cation up date messages to the HAs can b e reduced b y setting up a hierarc h y of foreign agen ts for mobilit y managemen t, where the lev el n um b er of the hierarc h y is dynamically adjusted based on eac h mobile user's up-to-date mobilit y and trac load information. Analytical mo del is dev elop ed for the p erformance ev aluation. Analytical results sho w that our new sc heme outp erforms the Mobile IP and IETF hierarc hical Mobile IP sc hemes under v arious conditions. The imp ortan t con tribution of our researc h is the new approac h w e dev elop in this w ork. Most p erformance ev aluation of mobilit y managemen t sc hemes in mobile IP net w orks is carried out b y sim ulations. Our w ork presen ts a no v el analytical approac h to the p erformance ev aluation of mobile IP net w orks. This c hapter is organized as follo ws. W e in tro duce the related w orks on the mobilit y managemen t for the Mobile IP net w orks in section 4.2; The detail pro cedures of the DHMIP sc heme are presen ted in section 4.3; In section 4.4, w e dev elop an analytical mo del to deriv e the signaling cost functions for the new sc heme; The DHMIP sc heme p erformance is demonstrated in section 4.5. Section 4.6 compares our new sc heme p erformance with that of the IETF Hierarc h y Mobile IP sc heme and some impro v emen ts whic h can enhance the DHMIP p erformance are detailed in section 4.7. Section 4.8 giv es the conclusions.

PAGE 72

61 4.2 Related W orks User mobilit y in wireless net w orks that supp ort IP mobilit y can b e broadly classied in to macro-mobilit y and micro-mobilit y The macro-mobilit y is for the case that an MH roams across dieren t administrativ e domains of geographical regions. The macro-mobilit y o ccurs less frequen tly and usually in v olv es longer timescales [12]. The Mobile IP can ensure the mobile users reestablish comm unication connections after a mo v e during macro-mobilit y The micro-mobilit y means the MH mo v emen t across m ultiple subnets within a single net w ork of domain. F or micro-mobilit y whic h o ccurs quite often, the Mobile IP paradigm needs to b e enhanced. Most of the related w orks attempt to impro v e the Mobile IP micro-mobilit y handling capabilit y [13]. In [66], the authors prop osed a scalable mobilit y managemen t sc heme whic h uses hierarc hical F As to handle the user mobilit y within one subnet w ork for wireless in ternet, and F A hierarc h y in this sc heme is pre-congured. In this arc hitecture, the Base Stations are assumed to b e net w ork routers. The higher lev els of the hierarc h y rely on Mobile IP to handle the macro-mobilit y The Hando-Aw are Wireless Access In ternet Infrastructure (HA W AI I) is a separate routing proto col to handle micro-mobilit y [40]. The sc heme hinges on the assumption that most user mobilit y is lo cal to an administrativ e domain of the net w orks. An MH en tering a new foreign net w ork is assigned a new CoA and retains its CoA unc hanged while mo ving within the foreign domain. In this sc heme, the HA and an y corresp onding host are una w are of the host's mobilit y within that domain. The route to the MH is established b y sp ecialized path setup sc hemes that up date the forw arding tables with host-based en tries in selected routes in that domain. F or macro-mobilit y the HA W AI I uses the traditional Mobile IP In this sense, this sc heme can b e considered as an enhanced Mobile IP The Cellular IP sc heme is in tro duced in [41]. In this sc heme, the lo cation managemen t and hando supp ort are in tegrated with routing in Cellular IP access net w orks. The net w ork is connected to the In ternet through a gatew a y router and the roaming b et w een gatew a ys is managed b y Mobile IP while the mobilit y within access net w orks is handled b y Cellular IP All the pac k ets originated from or terminated to the MHs are handled b y the gatew a ys, and the host lo cation information is refreshed b y the regular data exc hange transmitted from mobile host. The Cellular IP also supp orts IP paging. The P aging mec hanism can minimize the signaling trac supp orting the mobilit y managemen t for users in the idle state. When

PAGE 73

62 pac k ets need to b e sen t to an idle mobile host, the host is paged and b ecomes activ e. T o lo calize the signaling trac of supp orting IP services in cellular net w orks, Das et al [12] prop osed to use the mobilit y agen t (MA) to lo calize the up date trac, leading to a new arc hitecture, called T eleMIP In the T eleMIP an MA is in c harge of one region, handling the CoA addresses for those MHs roaming in the region. In the HA, the MA is registered. Whenev er an MH c hanges an F A to another F A in the same region, it only up dates the MA. When an MH crosses the region b oundary the MH registers with the new MA, whic h then sends an up date message to the HA. It has b een demonstrated that T eleMIP do es enhance the p erformance in the IP supp ort o v er the cellular net w orks. Ho w ev er, this new arc hitecture ma y add the net w ork managemen t en tit y and complexit y In the Mobile IP net w orks, to reduce the registration signaling trac (for binding up date), the Mobile IP regional registration is prop osed in [42], whic h is called IETF hierarc hical sc heme in this c hapter. The proto col emplo ys the F A hierarc h y to lo calize the registration trac. In this proto col, the HA registers the publicly routable address of the Gatew a y F A (GF A) and the MHs lo cation up date messages establish tunnels in a regional net w ork along the path from MHs to GF A. Although m ultiple lev els of hierarc h y are men tioned in [42], t ypically one lev el arc hitecture, where all F As are connected to the GF As, is used for discussions. In the IETF hierarc hical sc heme, the net w ork arc hitecture is cen tralized. It is not clear and usually hard to determine the size of a regional net w ork. The mobile users up-to-date trac load and mobilit y ma y v ary and the xed structure is lac k of rexibilit y T o o v ercome this deciency Xie and Akyildiz prop osed a distributed dynamic regional lo cation managemen t sc heme for Mobile IP [43]. In this sc heme, the rst F A an MH registers at in a new regional net w ork is selected as the GF A, and the regional net w ork size is adjusted based on the user's curren t trac load and mobilit y information. It can b e considered as the extension of the IETF regional registration sc heme to mak e it more rexible and adaptiv e. In this c hapter, w e prop ose another dynamic hierarc hical mobilit y managemen t sc heme for the Mobile IP net w orks. In our sc heme, when an MH c hanges its subnet and obtains a new care-of address from the new F A, the new F A up dates the new address to the MH's previous F A so that the new F A forms a new lo cation managemen t hierarc h y lev el for that user. The F A hierarc hical arc hitecture is sp ecic for ev ery user, whic h mak es the user a v oid up dating his/her home

PAGE 74

63 net w ork frequen tly The pac k ets deliv ered to the MH can b e tunnelled via the m ultiple lev els of F As to the user. In order to a v oid long pac k et deliv ery dela y there is an optimal n um b er (or threshold) of hierarc h y lev el for eac h user according to his/her call-to-mobilit y ratio ( C M R ). The threshold can b e dynamically adjusted based on the up-to-date mobilit y and trac load for eac h terminal. When the threshold is reac hed, the MH up dates its lo cation to the home net w ork and sets up a new hierarc h y for its further mo v emen ts. The optimal threshold for eac h user can b e deriv ed b y an iterativ e algorithm. Since our sc heme reduces the registration trac to the HAs signican tly it can also reduce the in-righ t pac k et loss greatly [68]. One signican t con tribution in this w ork is the analytical approac h w e dev elop for the p erformance ev aluation of wireless mobile IP net w orks. Most of past researc h fo cuses on the system arc hitecture for mobilit y managemen t in mobile IP net w orks. P erformance ev aluations are carried out mostly b y sim ulations. In our w ork, w e dev elop a systematically analytical approac h to ev aluate the prop osed mobilit y managemen t sc heme in mobile IP net w orks. It is our hop e that our w ork can op en a new a v en ue for p erformance ev aluation of mobile IP net w orks. 4.3 Dynamic Hierarc hical Lo cation Managemen t Sc heme In our new dynamic hierarc hical system arc hitecture, there is no xed hierarc hical arc hitecture for users or an y restriction on the shap e and the geographic lo cation of subnets. In the Mobile IP proto col, an MH can determine if it en ters a new subnet b y detecting the agen t adv ertisemen t messages sen t b y the mobilit y agen ts (HAs or F As). The MH then obtains a new CoA from the new serving F A and sends the lo cation up date message to its HA. Up on receiving the message, the HA can set up a binding b et w een the MH p ermanen t address and curren t CoA so that the HA can in tercept the pac k ets to this MH and tunnel them to the user's curren t access p oin t. The MHs in the Mobile IP net w orks are required to up date their new care-of addresses whenev er they c hange the lo cations (subnets) ev en though the MHs do not comm unicate with others. As sho wn in Fig. 4.1, this pro cedure could result in hea vy signaling trac to the net w orks. In our dynamic hierarc hical Mobile IP sc heme (DHMIP), the lo cation up date signaling trac can b e reduced b y registering the new CoA to the p ervious F A as sho wn in Fig. 4.2.

PAGE 75

64 FA 1 FA 2 FA 3 FA 4 FA 5 FA 6Location registration Packet delivery MH movement HA Home Agent FA Foreigen Agent MH Mobile Host k iHA MH MH MH INTERNET Figure 4.2: The DHMIP lo cation registration and pac k et deliv ery By this pro cedure, a dynamic lo cation hierarc h y is constructed for a sp ecic mobile user. The pac k ets for this user can b e in tercepted and retunneled along the F A hierarc h y to the mobile terminal. Th us, the lo cation up date trac can b e lo calized. In this sc heme, w e can adopt the similar pro cedures in [70, 71] to notify the previous F A the user's new CoA. Ho w ev er, the pac k et forw arding b y m ultiple F As will cause some service deliv ery dela y whic h ma y not b e appropriate when there is dela y restrain t for some in ternet applications suc h as video or v oice services. In order to a v oid excessiv e pac k et transmission dela y w e set a threshold to the hierarc h y lev el n um b er in the DHMIP sc heme. When the threshold is reac hed, the MH will register to its home agen t. In the DHMIP sc heme, the threshold is adjusted dynamically based on ev ery user's curren t trac load and mobilit y The p erformance of the DHMIP sc heme is sho wn in Fig. 4.2. In this gure, an MH mo v es from subnet 1 to subnet 6 W e assume the threshold lev el of the hierarc h y is three. When the user is in subnet 2 subnet 3 subnet 5 or subnet 6 the MH up dates the new care-of addresses to the previous F As. Since the previous F As are usually close to the new ones, the lo cation up date cost is less than that to HA. After the user en ters subnet 4 the hierarc h y lev el threshold is reac hed and the MH will set up a new hierarc h y In this case, the MH up dates its new CoA to the HA directly When the user is in subnet 3 or subnet 6 there are pac k ets arriv als for the user. The pac k ets are then in tercepted b y the HA and tunnelled to the user. Since the HA do es not

PAGE 76

65 T able 4.1: The dynamic hierarc hical Mobile IP proto col % Lo cation registration pro cedures K opt : the optimal hierarc h y lev el threshold; Initialize i = 0; if (MH en ters a new subnet) i = i + 1; if ( i < K opt ) New F A registers to the previous F A; else New F A registers to the HA; i = 0; Compute the new K opt ; % P ac k et deliv ery pro cedures if (pac k ets for the MH are in tercepted b y the HA) T unnel the pac k ets to the rst F A; if (the rst F A is not the MH curren t serving F A) Retunnel the pac k ets to the curren t F A; The curren t F A decapsulates the pac k ets and sends them to the MH; ha v e the user's up-to-date lo cation information, the pac k ets are sen t to the F A that the user up dated last time. In our example, they are F A 1 and F A 4 in Fig. 4.2. Th us, the pac k ets are retunneled along the hierarc h y to the user. The optimal hierarc h y lev el threshold K opt can b e computed based on the user's curren t trac load and mobilit y pattern. The K opt can b e adjusted in dieren t ep o c hs in the DHMIP sc heme. F or example, the optimal v alue can b e up dated ev ery time the MH en ters a new subnet or when the previously calculated threshold is reac hed or the MH calculates it p erio dically There is a trade-o b et w een the accuracy of K opt and the MH's computational p o w er consumption. The more often the up date of K opt the more accurate the v alue and the more signaling trac sa ving; ho w ev er the more p o w er consumption. The DHMIP sc heme can b e describ ed b y the pseudo-co de in T able 4.1. In this c hapter, w e assume the optimal v alue is up dated ev ery time the last optimal threshold is reac hed. 4.4 Analytical Mo del In this section, w e dev elop an analytical mo del to deriv e the lo cation up date and pac k et deliv ery cost functions for the DHMIP sc heme. Just lik e in [43], w e do not consider the p erio dic binding up date costs that an MH sends to its HA or F As to refresh their cac hes. W e dene the follo wing parameters for our analysis in the rest of this c hapter. : the crossed subnet n um b er b et w een the last pac k et arriv al and the nal lo cation up date just b efore it (see Fig. 4.2);

PAGE 77

66 : the MH call-to-mobilit y ratio ( C M R ); K : the threshold of the F A hierarc h y lev el; U : the a v erage MH lo cation up date cost to its HA; F : the pac k et deliv ery cost for a pac k et to MH in a foreign net w ork under the MIP sc heme; M 0 : the total lo cation up date cost for an MH incurred b et w een t w o consecutiv e pac k et arriv als under the DHMIP sc heme; F 0 : the a v erage pac k et deliv ery cost for a pac k et to MH in a foreign net w orks under the DHMIP sc heme; T : the pac k et deliv ery cost b et w een F As in the DHMIP sc heme; S : the hierarc h y setup cost in the DHMIP sc heme. Let ( i ) denote the probabilit y that an MH crosses i subnets b et w een t w o consecutiv e pac k et arriv als. Under the curren t Mobile IP sc heme, the total lo cation up date and pac k et deliv ery cost can b e sho wn as: C ( ) = 1 X i =0 iU ( i ) + F = U + F : (4.1) In this c hapter, the C M R is dened as follo w: if pac k ets arriv e at an MH at rate and time the user resides in a giv en subnet has a mean 1 = then the C M R denoted b y is giv en as = =: (4.2) The cost for the DHMIP is more complicated, so w e deriv e the lo cation up date and pac k et deliv ery costs, resp ectiv ely By the new sc heme, if w e assume the threshold is K as w e can see in Fig. 4.2, and if an MH crosses i subnets b et w een t w o consecutiv e pac k et arriv als, the MH will up date to the HA b i + K c times and up date to the p ervious F A in the rest i b i + K c times [45]. So the a v erage lo cation up date cost function can b e written as M 0 ( ; K ; ) = 1 X i =0 ( b i + K c U + ( i b i + K c ) S ) ( i ) : (4.3) Similarly w e can obtain the pac k et deliv ery cost function. In the DHMIP sc heme, some additional cost is in tro duced. When a pac k et is tunnelled to the rst F A in the MH curren t F A hierarc h y if the rst F A is not the user's serving F A, then additional F As ha v e to b e

PAGE 78

67 tra v ersed b efore the pac k et can reac h the destination. In addition to the pac k et deliv ery cost in the Mobile IP sc heme, there is additional ( i + b i + K c K ) T cost in the DHMIP sc heme. So the F 0 can b e obtained as follo ws: F 0 ( ; K ; ) = F + 1 X i =0 ( i + b i + K c K ) T ( i ) : (4.4) The ( i ) can b e express as [48] ( i ) = 8><>: 1 1 g if i = 0 (1 g ) 2 g i 1 if i > 0 (4.5) where g = f m ( s ), the Laplace transform of the residence time random v ariable densit y function. In order to analyze (4.3) and (4.4), w e assume that i = j K + q then ( j K + q ) = (1 g ) 2 g ( g K ) j g q = y z j x q ; (4.6) where y = (1 g ) 2 g ; z = g K ; x = g : Notice that b oth 0 q < K and 0 < K w e can rewrite M 0 as M 0 ( ; K ; ) = S 1 X i =0 i ( i ) + ( U S ) 1 X i =0 b i + K c ( i ) = S + ( U S ) 1 X j =0 K 1 X q =0 b j K + q + K c ( j K + q ) = S + ( U S ) 1 X j =0 K 1 X q =0 b j K + q + K c y z j x q = S + ( U S ) y 1 X j =0 [ j z j K 1 X q =0 x q + ( j + 1) z j K 1+ X q = K x q ] = S + ( U S ) y (1 x K + ) 1 x 1 X j =0 j z j + ( U S ) y ( x K x K + ) 1 x 1 X j =0 z j = S + ( U S ) y (1 x K + ) z (1 x )(1 z ) 2 + ( U S ) y ( x K x K + ) (1 x )(1 z ) = S + ( U S )(1 g ) g K 1 (1 g K ) [ 1 g K + 1 g K + 1 g 2 g ] : (4.7)

PAGE 79

68 Similarly w e can obtain F 0 ( ; K ; ) = F + 1 X i =0 iT ( i ) + 1 X i =0 T ( i ) K T 1 X i =0 b i + K c ( i ) = F + T + T K T (1 g ) g K 1 (1 g K ) [ 1 g K + 1 g K + 1 g 2 g ] : (4.8) Th us, the total cost is C 0 ( ; K ; ) = M 0 ( ; K ; ) + F 0 ( ; K ; ) = F + S + T + T + ( U S K T )(1 g ) g K 1 (1 g K ) [ 1 g K + 1 g K + 1 g 2 g ] : (4.9) F urthermore, if w e assume that the is uniformly distributed with = 1 K w e can rewrite (4.9) as C 0 unif or m ( K ; ) = 1 K K 1 X =0 C 0 = F + S + T + ( K 1) T 2 + ( U S K T )(1 g ) g K 1 (1 g K ) K K 1 X =0 [ 1 g K + 1 g K + 1 g 2 g ] = F + S + T + ( K 1) T 2 + ( U S K T )(1 g ) g K 1 (1 g K ) K [ K 1 g K + g 1 K g 1 1 g ] : (4.10) F or demonstration purp oses, w e assume that the mobile user subnet residence time is Gamma distributed with mean 1 = The Laplace transform of a Gamma distribution is f m ( s ) = r s + r r ; th us, w e ha v e g = f m ( ) = r + r r = r p + r r :

PAGE 80

69 In particular, when r = 1, w e ha v e an exp onen tial distribution for the subnet residence time. If the residence time is exp onen tially distributed, w e ha v e g = 1 1+ then (4.10) is reduced to C 0 unif or m ( K ; ) = F + S + T + ( K 1) T 2 + U S K T (1 + ) K 1 [ (1 + ) K (1 + ) K 1 + (1 + ) K 2 K ] : (4.11) In realit y man y mobile users usually k eep accessing the In ternet in a vicinit y When they roam far a w a y from their daily w ork place, they w ould access the net w ork less often. So w e can sim ulate the situations b y setting linearly or exp onen tially distributed. If is linearly distributed with = 2( K ) K ( K +1) then C 0 l inear ( K ; ) = K 1 X =0 C 0 = F + S + T + ( K 1) T 3 + ( U S K T )(1 g ) g K 1 (1 g K ) K 1 X =0 2( K ) K ( K + 1) [ 1 g K + 1 g K + 1 g 2 g ] : (4.12) If is exp onen tially distributed with = e (1 e 1 ) 1 e K then C 0 exponential ( K ; ) = K 1 X =0 C 0 = F + S + T + T [ e 1 + ( K 1) e K 1 K e K (1 e K )(1 e 1 ) ] + ( U S K T )(1 g ) g K 1 (1 g K ) K 1 X =0 e (1 e 1 ) 1 e K [ 1 g K + 1 g K + 1 g 2 g ] : (4.13) F or MHs with dieren t trac load and mobilit y patterns, their optimal hierarc h y lev el thresholds should b e dieren t. The optimal threshold ( K opt ) for an MH is the v alue of K that minimizes the cost functions deriv ed ab o v e. Since the K opt m ust b e an in teger, w e use a similar metho d in [31, 43] to obtain the optimal v alues. W e dene the cost dierence

PAGE 81

70 equation b et w een the system with lev el K and the one with lev el K 1 ( K 2) as ( K ; ) = C 0 ( K ; ) C 0 ( K 1 ; ) : (4.14) Giv en the C M R the algorithm to nd the optimal v alue of K is dened as K opt = 8><>: 1 if (2 ; ) > 0 max f K : ( K ; ) 0 g otherwise (4.15) Notice that the algorithm to get the optimal v alue is iterativ e. It is easy to implemen t; ho w ev er it ma y result in lo cal minima. Ho w to a v oid the lo cal minima w as discussed in [69]. In practical systems, w e can determine the optimal v alue to a limited predened maxim um n um b er. Then, the optimal v alue can b e found b y ev aluating the total costs for eac h of the allo w ed n um b ers in the range. The estimation of the MH pac k et arriv al rate w as also discussed in [31], whic h w e will use in this w ork. 4.5 Numerical Results T o ev aluate the p erformance of the DHMIP sc heme, w e need to kno w the relativ e op eration costs. W e use the follo wing notations for our analysis. m mf u : the transmission cost of lo cation up date b et w een an MH and an F A; m f f u : the transmission cost of lo cation up date b et w een F As; m f h u : the transmission cost of lo cation up date b et w een an F A and the HA; p f h u : the lo cation up date pro cessing cost b et w een an F A and the HA; p mf u : the lo cation up date pro cessing cost b et w een an MH and an F A; p f f u : the lo cation up date pro cessing cost b et w een F As; m f f d : the transmission cost of pac k et deliv ery b et w een F As; m hf d : the transmission cost of pac k et deliv ery b et w een the HA and an F A; m f m d : the transmission cost of pac k et deliv ery b et w een an F A and an MH; p f f d : the pac k et deliv ery pro cessing cost b et w een F As; p hf d : the pac k et deliv ery pro cessing cost b et w een the HA and an F A; p f m d : the pac k et deliv ery pro cessing cost b et w een an F A and an MH. According to the ab o v e denitions and the proto cols of the Mobile IP and the DHMIP sc hemes describ ed in Section 4.3, the MH registration cost ( U ) to the HA, the pac k et deliv ery cost ( F ) in the Mobile IP net w ork, the hierarc h y setup cost ( S ) and the hierarc hical pac k et

PAGE 82

71 T able 4.2: The p erformance analysis parameters (1) p f m d p f f d p hf d m f m d m f f d 1 0.5 10 5 25 p mf u p f f u p f h u m mf u m f f u 2 1 20 10 50 T able 4.3: The p erformance analysis parameters (2) set m f h u m hf d 1 250 125 2 500 250 3 1000 500 deliv ery cost ( T ) in the DHMIP can b e expressed as: U = 2 m mf u + 2 m f h u + 2 p mf u + p f h u ; (4.16) S = 2 m mf u + 2 m f f u + 2 p mf u + p f f u ; (4.17) T = m f f d + p f f d ; (4.18) F = m hf d + m f m d + p hf d + p f m d : (4.19) The Mobile IP sc heme ma y result in hea vy system trac load b ecause of the MHs' lo cation up date messages. The DHMIP sc heme attempts to reduce the registrations to the HAs b y informing MHs' new lo cation to the old F As. This w ould eectiv ely reduce the long distance signaling trac at the exp ense of increasing the lo cal trac load. Usually an MH's new F A is close to its old one in term of the hops b et w een them and the trac load is distributed ev enly among the system, so that the net w ork can accept more service requests. It is ob vious that the DHMIP sc heme p erformance dep ends on the relativ e cost of the long distance registration to the HA to the lo cal registration to F As. In this c hapter, w e use the parameters in T able 4.2 and T able 4.3 for p erformance analysis. W e use three sets of parameters in our analysis. The signaling costs b et w een the adjacen t hierarc h y lev els are xed and the costs b et w een the HA and an F A are 5, 10 and 20 times of those in the three sets of parameters, resp ectiv ely Fig. 4.3 sho ws the total signaling costs for Mobile IP and the DHMIP sc hemes under v arious C M R v alues. In this gure, w e observ e that when the C M R is small, the DHMIP sc heme generates m uc h less trac than the Mobile IP sc heme do es ev en without using the

PAGE 83

72 10 1 10 0 10 1 10 2 10 3 10 4 CalltoMobility Ratio ( r )The total cost MIPDHMIP with K=4DHMIP with K opt Figure 4.3: Comparison of the total costs for dieren t sc hemes optimal v alues. W e also see that our sc heme can reduce the trac load ev en further with the optimal v alues. Ho w to obtain the optimal v alue has b een discussed in Section 4.4. If w e examine Fig. 4.3 more carefully w e can see that the total cost for the DHMIP sc heme with xed threshold exceeds that for the Mobile IP sc heme when the C M R is large. This can b e seen more clearly in Fig. 4.4. In this gure, w e sho w the relativ e costs for the DHMIP sc heme to that for Mobile IP sc heme under dieren t C M R The uniform distribution is used and the thresholds are four for all the curv es in this gure. It is ob vious that the total cost for the DHMIP sc heme with xed threshold can exceed that for the Mobile IP sc heme when the C M R is large. In the new sc heme, the pac k ets for an MH will b e pro cessed and tunnelled through more F As b efore it reac hes the destination. When the pac k et arriv al rate is high comparing with the mobilit y and the threshold is xed, the total cost ma y exceed that for the Mobile IP sc heme. Ho w ev er, w e can sho w that, with the optimal v alues, the DHMIP sc heme will nev er generate more trac than the Mobile IP sc heme do es under an y conditions. W e can also see in Fig. 4.4 that the sc heme p erforms b etter when the relativ e cost to the HA is higher, the reason is straigh tforw ard. Some mobile users usually roam only in a vicinit y when they access the in ternet applications. F or example, a passenger w aiting for his/her righ t in terminal or the p eople in some construction eld. They usually require less service when they mo v e a w a y from some

PAGE 84

73 10 1 10 0 10 1 10 2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 The total cost ratio (C'/C)CalltoMobility Ratio ( r ) Set 3Set 2Set 1 Figure 4.4: Comparison of the total costs for uniformly distributed under dieren t parameterssp ecic areas. In this c hapter, w e use the linear and exp onen tial distributions to sim ulate those kinds of mobile users and sho w the p erformance in Fig. 4.5. It can b e seen that, with xed threshold, the users roaming around a sp ecic area can generate less signaling trac under the DHMIP sc heme. In fact, when a mobile user roams in a sp ecic area, there is high probabilit y that the user will revisit some subnets. This will generate lo ops in the hierarc h y and result in signaling trac whic h can b e remo v ed b y \lo op remo v al" in tro duced in Section 4.7. Before discuss the c hoice of the optimal threshold for a mobile user, w e study the eect of the hierarc h y lev el n um b er on the DHMIP sc heme p erformance rst. Under v arious C M R v alues, w e c hange the n um b er of hierarc h y lev els K in Fig. 4.6. When the C M R is lo w, meaning that the user has relativ ely high mobilit y larger K can minimize the total signaling costs b ecause more up dates to the HA can b e replaced b y hierarc h y setup, whic h will cost less. Ho w ev er, when the C M R is high, the pac k ets deliv ered to an MH ha v e to tra v el more F As b efore reac hing the destination with larger K In this case, the total costs of the DHMIP sc heme ma y exceed those of the MIP sc heme. W e can also see in Fig. 4.6 that it is hard to nd an optimal xed threshold for an MH under dieren t situations. In order to ac hiev e the b est p erformance, the selection of threshold m ust b e dynamic.

PAGE 85

74 10 1 10 0 10 1 10 2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 The total cost ratio (C'/C)CalltoMobility Ratio ( r ) uniformlinearexponential Figure 4.5: Comparison of the total costs under dieren t distributions 10 1 10 0 10 1 10 2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 The total cost ratio (C'/C)CalltoMobility Ratio ( r ) K=2K=4K=6 Figure 4.6: Comparison of the total costs under dieren t K

PAGE 86

75 10 1 10 0 10 1 10 2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 The total cost ratio (C'/C)CalltoMobility Ratio ( r ) uniformlinearexponential Figure 4.7: Comparison of the total costs under dieren t distributions with optimal K 1 0.5 0 0.5 1 1.5 2 0 5 10 15 20 25 30 35 40 45 Optimal KCalltoMobility Ratio lg10( r ) set 3set 2set 1 Figure 4.8: The optimal n um b er of hierarc h y lev els for uniform distribution

PAGE 87

76 10 2 10 0 10 2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 rTotal cost ratio for uniform distribution g =00.1 g =1 g =100 10 2 10 0 10 2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 rTotal cost ratio for linear distribution g =00.1 g =1 g =100 10 2 10 0 10 2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 rTotal cost ratio for exponential distribution g =00.1 g =1 g =100 Figure 4.9: The eect of the v ariance of the F A residence time ( r ) In Fig. 4.7, w e sho w the relativ e p erformance for the DHMIP sc heme using the optimal v alues under dieren t C M R v alues and dieren t distributions. W e can observ e that our new sc heme can reduce 85% of the signaling cost when the C M R is small. With the increase of C M R the relativ e cost also increases. Ho w ev er, the cost for the DHMIP sc heme will nev er exceed the cost for the MIP sc heme. Fig. 4.8 sho ws the optimal v alues under dieren t C M R for the three sets of parameter. In this gure, w e assume that the for the MH follo ws a uniform distribution. In Fig. 4.8, the optimal v alues decrease with the increase of C M R When the relativ e up date cost to the HA is large, an MH can ha v e relativ ely large thresholds. The reason is ob vious. If the up date cost to the HA is high, an MH can generate less signaling trac b y setting up more F A hierarc h y otherwise, the MH should up date the HA more often, whic h implies smaller K In fact, when the optimal v alue equals one, our new sc heme b ecomes the MIP sc heme, so that the total costs will nev er exceed those for the MIP sc heme. Based on the ab o v e analysis, w e can see that the MIP sc heme is suitable for mobile users with high C M R and our DHMIP sc heme can p erform w ell for all kinds of users. W e next in v estigate the sensitivit y of the p erformance costs and the b enets of the DHMIP sc heme to the v ariance of the mobile user's mobilit y pattern. In our analysis, w e assume that the pac k et arriv als to a mobile user form a P oisson pro cess and the user's

PAGE 88

77 residence time in a subnet has a Gamma distribution. F or a Gamma distribution, the v ariance is V = 2 r That is, a large r implies a small v ariance. Fig. 4.9 demonstrates the eects of the v ariance of the F A residence time under three distributions. In this gure, w e assume r = 0 : 01, 1 and 100, resp ectiv ely W e can see in the gure that, the v ariance of the F A residence time in the subnets do es not aect the p erformance of the DHMIP sc heme v ery m uc h. W e deriv e the optimal v alues based on the exp onen tial residence time distribution. When the r = 0 : 01 or 100, those v alues are not optimal an ymore. Ho w ev er, the DHMIP can still ac hiev e similar p erformance under dieren t C M R dieren t residence time and distributions. This phenomenon pro v es the rexibilit y and the adaptivit y of our new sc heme. 4.6 Comparison with the IETF Hierarc hical Sc heme The IETF Hierarc hical Mobile IP (HMIP) proto col aims to reduce the signaling trac to the home net w orks while reducing the signaling dela y The sp ecication of the HMIP proto col can b e found in [42]. In this proto col, the HMIP emplo ys a hierarc h y of F As to lo cally handle Mobile IP registration. When an MH mo v es from one subnet to another, it p erforms a home registration to its HA. During a home registration, the HA registers the care-of address of the MH. When the visited domain supp orts regional tunnel managemen t, the care-of address that is registered at the HA is the publicly routable address of a Gatew a y F oreign Agen t (GF A). This care-of address will not c hange when the mobile no de c hanges foreign agen t under the same GF A. In this situation, the MH will p erform a regional up date to GF A, the MH mobilit y managemen t can b e handled lo cally in this w a y During the comm unication, when pac k ets are sen t to the MH, the user's HA in tercepts the pac k ets, encapsulates them and tunnels them to the CoA of the MH. Those pac k ets will reac h the GF A of the MH through the net w ork. The GF A c hec ks its visitor list and re-tunnels the pac k ets to the user's corresp onding F A. The F A further rela ys the pac k ets to the MH. T ypically one lev el of hierarc h y where all F As are connected to the GF As, is considered; ho w ev er the proto col ma y b e utilized to supp ort m utiple lev els of hierarc h y as discussed in App endix B in [42]. In this section, w e compare our DHMIP sc heme with the IETF hierarc hical mobile IP sc heme. W e assume U 1 is the MH lo cation up date cost to a GF A, U 2 is that to the HA and F g f is the additional pac k et deliv ery cost induced b y the GF A. In the IETF HMIP sc heme,

PAGE 89

78 the system structure is xed. If an MH roams under the same GF A, the user total mobilit y managemen t costs can b e reduced. Ho w ev er, the HMIP proto col requires the MH to p erform home registration when user en ters a subnet c harged b y another GF A. In order to analyze the HMIP p erformance, w e need to kno w the probabilit y that a user mo v es out of a regional net w ork. W e further assume that is the probabilit y that the next mo v emen t of MH is under the same GF A. In order to obtain w e assume that there are N subnets in a foreign net w ork and m subnets in one regional area, resp ectiv ely In the mo del, w e dene the action that eac h MH mo ving out of a subnet a movement In one mo v emen t, an MH can visit a subnet randomly Then, w e can ha v e = m 1 N 1 Usually the distance b et w een a GF A and the MH's curren t F A is farther than the F As in our DHMIP sc heme in term of hops, so the lo cation up date and pac k et deliv ery costs are also higher. Conserv ativ ely w e dene the cost co ecien t = 2. The GF A pro cessing cost is prop ortional to the n um b er of subnets under it. Since IP routing table lo okup is based on the longest pr ex matching then the GF A pro cessing complexit y is prop ortional to the logarithm of m According to the ab o v e assumptions w e can ha v e U 1 = S l og ( m ), U 2 = U + U 1 and F g f = T l og ( m ). Th us, the total cost function for IETF HMIP sc heme can b e expressed as C 0 H M I P ( ) = 1 X i =0 i [ U 1 + (1 ) U 2 ] ( i ) + F + F g f = U 1 + (1 ) U 2 + F + F g f = U ( N m ) ( N 1) + S l og ( m ) + F + T l og ( m ) : (4.20) In Fig. 4.10, w e plot the relativ e cost curv es for the IETF HMIP sc heme and DHMIP sc heme to that of the basic MIP proto col. In this gure, w e assume N = 50, m = 10 and 20, resp ectiv ely W e can see in this gure that our DHMIP sc heme generates less system signaling cost than the IETF HMIP sc heme under all the conditions ev en without using the optimal threshold v alues. W e can also see that if the n um b er of subnets under one GF A is large, the HMIP sc heme can reduce the system signaling trac more when the C M R is small. Ho w ev er, the total cost increases rapidly with the increase of C M R and exceeds that with smaller m So, for the IETF HMIP sc heme, it is hard to nd an optimal n um b er of subnets under one GF A for all the users in one net w ork. Another dra wbac k of the HMIP

PAGE 90

79 101 100 101 102 0 0.2 0.4 0.6 0.8 1 1.2 CalltoMobility Ratio ( r )The total cost ratio (C'/C) DHMIPw =19/49 w =9/49 Non optimal Figure 4.10: The comparison with the IETF hierarc hical sc heme sc heme, as men tioned in [43], is that the cen tralized system arc hitecture mak es the system p erformance sensitiv e to the failure of GF As. Our DHMIP sc heme a v oids this problem successfully 4.7 Impro v emen t of DHMIP Sc heme The p erformance of the DHMIP sc heme can b e impro v ed further b y the state activation and lo op r emoval 4.7.1 State Activ ation Curren tly Mobile IP supp orts registrations but not paging. IP paging is an IP la y er pro cess for an IP net w ork to determine an MH's precise lo cation, i.e., the precise attac hmen t p oin t to the net w ork where it can receiv e IP pac k ets. With the IP paging supp ort, MHs can reduce the registration to the HA when they do not engage an y comm unication. In the IP paging proto col, the HA only k eeps appro ximate lo cation information of MH whic h is in idle state. The MH will turn to activ e state when pac k ets are receiv ed and up date its curren t lo cation to the HA. W e can enhance the p erformance of the DHMIP sc heme b y setting up a long threshold for MH in idle state in a similar w a y In this w ork, w e name the impro v ed DHMIP sc heme Enhanc e d DHMIP scheme Similar to the Cellular IP [41], w e dene the id le MH as one that has not receiv ed data pac k et for a system sp ecic p erio d. In the Enhanced DHMIP sc heme, the MH can k eep a xed relativ ely large threshold K so

PAGE 91

80 that the MH can a v oid up dating the home net w ork often. An idle mobile host that receiv es the rst pac k et mo v es from idle to activ e state and up dates its curren t lo cation to the HA immediately and k eeps doing it when it en ters new subnets if it k eeps in activ e state. In the Enhanced DHMIP sc heme, the data pac k ets can a v oid the extra transmission dela y in the DHMIP sc heme and the total signaling trac is reduced at the same time. If the mobile user has not receiv ed pac k et for some predened p erio d, the MH returns bac k to the idle state again. In this enhanced v ersion, an MH can main tain a xed threshold K without computing and c hanging it from time to time. This can also reduce the mobile terminal energy consumption. F or dieren t mobile users with dieren t call to mobilit y patterns, they can set dieren t predened threshold v alues. In our analysis, w e assume that the MH turns to activ e state after receiving the rst data pac k et and can up date to its home net w ork b y attac hing the lo cation information to the ac kno wledge messages. Then, w e can obtain the lo cation up date cost function ( E M 0 ), pac k et deliv ery cost function ( E F 0 ) and the total cost function ( E C 0 ) as follo ws: E M 0 ( K ; ) = 1 X i =0 [ b i K c U + ( i b i K c ) S ] ( i ) ; = S + (1 g ) g K 1 ( U S ) (1 g K ) ; (4.21) E F 0 ( K ; ) = F + 1 X i =0 ( i K b i K c ) T ( i ) = F + [1 K g K 1 + ( K 1) g K ] T (1 g K ) ; (4.22) E C 0 ( K ; ) = E M 0 ( K ; ) + E F 0 ( K ; ) = F + S + ( U S )(1 g ) g K 1 + [1 K g K 1 + ( K 1) g K ] T (1 g K ) = F + S + T + ( U S K T )(1 g ) g K 1 (1 g K ) : (4.23) W e further assume that the MH's F A residence time in subnets is exp onen tially distributed, then g = 1 1+ w e can rewrite (4.23) as E C 0 ( K ; ) = F + T + S + U S K T (1 + ) K 1 : (4.24)

PAGE 92

81 10 1 10 0 10 1 10 2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 The total cost ratio (C'/C)CalltoMobility Ratio ( r ) K=2K=4K=10 Figure 4.11: The total cost for the enhanced DHMIP sc heme Fig. 4.11 sho ws the p erformance of the Enhanced DHMIP sc heme. W e can observ e from the gure that there is no optimal threshold for an MH. The larger the threshold, the b etter the p erformance. Ho w ev er, a large threshold ma y generate long pac k et deliv ery dela y for the rst data pac k et so that the net w ork op erator should c ho ose the v alue b y considering the signaling sa ving and the qualit y of services comprehensiv ely In Fig. 4.11, w e also see that the K has little eect on the sc heme p erformance when the C M R is large. The reason is that the MH is alw a ys in activ e state and k eeps up dating its newly obtained CoA to the HA in this situation and the eectiv e threshold reduces to one no matter what the predened v alue is. In the Enhanced DHMIP sc heme, the total system signaling cost will nev er exceed that of the MIP sc heme under all the conditions. 4.7.2 Lo op Remo v al The p erformance for the DHMIP sc heme can also b e impro v ed b y remo ving the lo ops formed during the mobile user's roaming. In realit y man y mobile users roam in a limited region, for example, in an oce building. If the mobile users revisit some subnets they ha v e visited b efore, lo ops ma y form in the hierarc hical arc hitecture in the DHMIP sc heme according to the proto col describ ed b efore. With the lo op r emoval the MHs can up date their new care-of addresses less often and the total signaling cost can b e minimized further [23]. When an MH en ters a new subnet and tries to set up a new lev el of hierarc h y the new F A

PAGE 93

82 c hec ks its hierarc h y list rst. If the F A is already in the MH hierarc h y the F A can delete the subsequen t F A addresses for that user without up dating the old F A so that the lo op is remo v ed. In this section, w e analyze the lo op r emoval eect on the DHMIP p erformance. If w e consider the MH revisiting scenario, w e can dene ( i ) as the eectiv e n um b er of subnets crossed b et w een t w o consecutiv e pac k et arriv als. Although the lo op r emoval can reduce the n um b er of HA up dates, the MH has to up date its lo cation to the curren t F A ev en the MH has visited that F A b efore. W e dene = 2 m mf u + 2 p mf u as the signaling cost of an MH notifying its arriv al to the curren t F A, w e can obtain the cost function with lo op r emoval as N C 0 ( ; K ; ) = K 1 X =0 1 X i =0 h b ( i ) + K c ( U ) + ( ( i ) b ( i ) + K c )( S ) + i + ( ( i ) + b ( i ) + K c K ) T i ( i ) + F = F + + T K 1 X =0 + ( U S K T ) K 1 X =0 1 X i =0 b ( i ) + K c ( i ) +( S + T ) 1 X i =0 ( i ) ( i ) : (4.25) F or simplicit y and demonstration purp ose, w e assume that is uniformly distributed and ( i ) = i where is the p ercen tage of the ee ctive n um b er of subnets a user visited to the total n um b er of subnets the user visited. Then, w e ha v e N C 0 ( K ; ) = F + ( K 1) T 2 + ( S + T ) + (1 ) + ( U S K T ) K K 1 X =0 1 X i =0 b i + K c ( i ) : (4.26) Fig. 4.12 sho ws the lo op r emoval eect when is 1, 0 : 8, 0 : 5 and 0 : 2, resp ectiv ely When = 1, it is the w orst case that the mobile user nev er revisits an y subnet he/her visited b efore. F rom Fig. 4.12, w e can see that the total signaling cost for the DHMIP sc heme can b e reduced b y the lo op r emoval mec hanism. 4.8 Conclusions This c hapter has presen ted a new lo cation managemen t sc heme for Mobile IP net w ork| the Dynamic Hierarc h y Mobile IP (DHMIP) strategy Instead of up dating the home net w orks

PAGE 94

83 10 1 10 0 10 1 10 2 0 0.2 0.4 0.6 0.8 1 1.2 The total cost ratio (C'/C)CalltoMobility Ratio ( r ) x =1 x =0.8 x =0.5 x =0.2 Figure 4.12: The total cost with lo op r emoval far a w a y in the DHMIP sc heme, the MHs inform their new care-of addresses to the previous F As. When data pac k ets arriv e at the MHs, the pac k ets can b e deliv ered through the F A hierarc h y The transmission distance b et w een the F As is usually shorter than that b et w een an MH and the HA, so the total signaling cost can b e reduced. In this c hapter, w e also prop osed an iterativ e algorithm to compute the optimal threshold v alues for sp ecic users with dieren t call to mobilit y patterns. Our analysis pro v es that the DHMIP sc heme outp erforms the traditional MIP sc heme under v arious conditions. In order to impro v e the MIP p erformance, the Hierarc hical Mobile IP (HMIP) proto col has b een prop osed b y IETF. In the IETF HMIP sc heme, the net w ork arc hitecture is cen tralized and sensitiv e to the failure of GF A. It is also pro v ed in our analysis that it is imp ossible to nd an optimal n um b er of subnets under one GF A for all kinds of users. W e compared the IETF HMIP sc heme with our DHMIP in this c hapter, the results sho w that the DHMIP sc heme can ac hiev e b etter p erformance. The DHMIP p erformance can also b e impro v ed b y considering the state activ ation and it lo op remo v al. All the analysis sho ws that the DHMIP can minimize the lo cation managemen t cost for Mobile IP net w orks and has more rexibilit y and robustness than the traditional Mobile IP sc heme.

PAGE 95

CHAPTER 5 CONCLUSIONS AND FUTURE W ORKS The rapid gro wth of wireless mobile net w orks and services, fuelled b y the next generation mobile comm unications systems researc h, has ushered in the era of ubiquitous computing. Ligh t w eigh t p ortable computers, IP-based appliances, and the p opularit y of the In ternet are pro viding strong incen tiv es to service pro viders to supp ort seamless user mobilit y It is exp ected that future wireless mobile net w orks will b e more heterogeneous and that ev ery mobile user will b e able to gain access to the In ternet bac kb one b y attac hing his or her mobile computing devices to a wireless access p oin t. Besides some new tec hnologies required to mak e the ab o v e scenarios realized, the user mobilit y managemen t proto cols are confron ted with more and more c hallenges. Mobilit y is an essen tial c haracteristic of the wireless comm unication net w orks. With the increase of the mobile user n um b er and the decrease of the registration area size, the lo cation managemen t trac will increase dramatically in the 3rd generation wireless comm unication signaling net w orks. The researc h aiming to minimize and optimize the mobilit y managemen t cost has b een carried out extensiv ely In this dissertation, w e prop osed t w o sc hemes based on p oin ter forw arding and one based on user mobilit y pattern for wireless comm unication net w orks. The p oin ter forw arding sc hemes include t w o-lev el p oin ter forw arding and POFLA sc hemes. In b oth sc hemes, w e try to impro v e the original mobilit y managemen t sc hemes b y in tro ducing a new functional en tit y|Mobilit y Agen t (MA). The MAs can either b e soft w arebased or hardw are-based and can b e distributed dynamically among the net w ork according to the users' curren t trac and mobilit y This c haracteristic mak es the new sc hemes more adaptiv e and robust. In b oth sc hemes, w e designed t w o kinds of p oin ters for dieren t c harge domains: high lev el p oin ter is set up when the threshold for lo w lev el p oin ter c hain is reac hed and home lo cation registration is p erformed after the high lev el p oin ter limit is reac hed. The t w o new sc hemes impro v e the net w ork mobilit y managemen t p erformance greatly comparing to some other prop osed sc hemes, suc h as p er-user forw arding and lo cation anc hor strategies. 84

PAGE 96

85 The t w o-lev el p oin ter forw arding and POFLA sc hemes a v oid the b ottlenec k eect in the anc hors for the lo cal anc hor sc heme and o v ercome the small p oin ter length limit for the p er-user forw arding strategy In our w orks, w e dev elop ed analytical mo dels for the wireless comm unication systems, analyzed and obtained the cost functions for the aforemen tioned proto cols. The results giv e a clear picture of the relativ e p erformance for all the sc hemes. Our new p oin ter forw arding sc hemes outp erform the original proto cols and some impro v ed v ersions under v arious conditions. Based on the observ ation that man y mobile users follo w some xed daily routines, w e prop osed another new mobilit y managemen t sc heme|user mobilit y patter based sc heme (MPBS). In this sc heme, the users' mobilit y patterns are stored in the user proles and hence the net w ork can determine the users' curren t lo cation. The MPBS sc heme cannot only reduce the system lo cation managemen t signaling trac but also mak e it p ossible for the service pro viders to pro vide users user-orien ted services. The system resources can b e allo cated more reasonably and ecien tly b ecause the user mobilit y pattern is included in the user prole and accurate prediction can b e made b y the system according to the information. W e v eried our results b y sim ulations in our w orks. All the results suggest that the MPBS sc heme can signican tly impro v e the net w ork mobilit y managemen t p erformance. The total cost for MPBS sc heme increases slo wly with the unit paging cost. It is an impressiv e adv an tage for the 3rd generation systems b ecause the radio transmission cost will k eep increasing in the future. The 3rd Generation P artnership Pro ject (3GPP) and 3rd Generation P artnership Pro ject 2 (3GPP2) are the consortium of in ternational standards b o dies task ed with dev eloping arc hitectures and standards for the third-generation cellular systems. Although more harmonization is needed b efore the nal standards b e accepted b y all the nations, it is no w widely recognized that using IP as the foundation for next-generation mobile net w orks mak es strong economic and tec hnical sense. T raditionally wide-area mobilit y o v er the IP net w orks has b een based on the family of Mobile IP proto cols. Ho w ev er, the micro-mobilit y of the Mobile IP p erformance needs to b e enhanced b ecause of the hea vy signaling trac and long latency In our researc h, w e extended the IETF regional registration proto col b y prop osing the Dynamic Hierarc h y Mobile IP lo cation managemen t sc heme (DHMIP). In the DHMIP sc heme, an MH up dates its new lo cation to the previous F A so that a dynamic hierarc h y is set up

PAGE 97

86 for the sp ecic user. Since the hierarc hies are distributed among the net w ork and adjusted based on the users' up-to-date trac burden and mobilit y the new sc heme can impro v e the IP net w ork mobilit y managemen t p erformance greatly The new DHMIP strategy also successfully a v oids the single failure p oin t problem in the IETF regional registration proto col so that the net w ork robustness and surviv abilit y are enhanced at the same time. The most imp ortan t con tribution of our researc h is the analytical approac h w e dev elop ed in this w ork. Almost all the previous w orks dealing with the IP net w ork mobilit y w ere carried b y sim ulations. In our researc h, w e dev elop ed a systematic mo del and deriv ed the close-form solution for the cost functions for the Mobile IP and DHMIP sc hemes. It is more straigh tforw ard to nd out ho w the net w ork conguration parameters can aect the proto col p erformance. W e hop e the new approac h w e dev elop ed can op en a new a v en ue to the future researc h in this area and the mo del can b e consummated b y other researc hers. In our researc h, when w e dev elop the system mo del and analyze the p erformances, w e alw a ys assume the P oisson and exp onen tial distributions for some v ariables. The reason w e use the ab o v e distributions is that they are easy to analyze so that w e can obtain analytical expressions. In realit y those assumptions ma y not b e realistic. Ho w to generalize our analytical results under more realistic assumption b ecomes an imp ortan t researc h task. Another issue ab out the DHMIP proto col is the user data arriv al pattern. With more and more kinds of services pro vided o v er the In ternet, esp ecially the real time data services, it is b ecoming more and more dicult to mo del the pac k et transmissions. In our mo del, if the pac k et arriv al can b e replaced b y the service session, the results could b e more p ersuasiv e. Ho w ev er, it is still an op en issue ab out the session mo del o v er the m ultimedia net w orks based on IP proto col.

PAGE 98

REFERENCES [1] I.F. Akyildiz, J. McNair, J.S.M. Ho, H. Uzunalioglu and W. W ang, \Mobilit y Managemen t in Next-Generation Wireless Systems," Pr o c. of the IEEE v ol.4, no.5, pp.13471384, Octob er 1996. [2] V.W.S. W ong and V. C. M. Leung, \Lo cation Managemen t for Next-Generation P ersonal Comm unications Net w orks," IEEE Network v ol.14, no.5, pp.18-24, Septemb er/Octob er 2000. [3] Theo dore S. Rappap ort, A. Annamalai, R. M. Buehrer and William H. T ran ter, \Wireless Comm unications: P ast Ev en ts and A F uture P ersp ectiv e," IEEE Communic ations Magazine v ol.40, no.5, pp.148-161, Ma y 2002. [4] S. T abbane, \Lo cation Managemen t Metho ds for Third-Generation Mobile System," IEEE Communic ation Magazine v ol.35, no.8, pp.72-84, August, 1997. [5] D. C. Co x, \Wireless P ersonal Comm unications: What Is It?" IEEE Personal Communic ations v ol.2, no.2, pp.20-35, April, 1995. [6] Y. B. Lin, \Determining the User Lo cations for P ersonal Comm unications Services Net w orks," IEEE T r ans. V eh. T e chnolo gy v ol.43, no.3, pp.466-473, Aug. 1994. [7] EIT/TIA, \Cellular Radio T elecomm unications In tersystem Op erations," T ec hnical Rep ort IS-41 (Revision B), EIA/TIA (1991). [8] ETSI, \Digital Cellular T elecomm unications System (Phase 2+): Mobile Application P art (MAP) Sp ecication (GSM 09.02 v ersion 7.51 Release)," 1998. [9] Y.B. Lin, Y.R. Haung, Y.K. Chen and I. Chlam tac, \Mobilit y Managemen t: from GPRS to UMTS," Wir eless Communic ations and Mobile Computing (WCMC) v ol.1, no.4, pp.339-359, 2001. [10] Y. B. Lin, \Reducing Lo cation Up date Cost in a PCS Net w ork," IEEE/A CM T r ans. on Networking v ol.5, no.1, pp.25-33, F eb. 1997. [11] Charles E. P erkins, \Mobile IP ," IEEE Communic ations Magazine v ol.5, no.2, pp.8499, Ma y 1997. [12] S. Das, A. Misra and P Agra w al, \T eleMIP: T elecomm unications-Enhanced Mobile IP Arc hitecture for F ast In tradomain Mobilit y ," IEEE Personal c ommunic ations v ol.7, no.4, pp.50-58, August, 2000. [13] Andrew T. Campb ell, Ja vier Gomez, Sangh y o Kim, Chieh-Yin W an, Zoltan R. T uran yi and Andras G. V alk o, \Comparison of IP Micromobilit y Proto cols," IEEE Wir eless Communic ations v ol.9, no.1, pp.72-82, F eburary 2002. [14] I. F. Akyildiz and J. S. M. Ho, \On Lo cation Managemen t for P ersonal Comm unications Net w orks," IEEE Comm. Mag. v ol.34, pp.138-145, Sept. 1996. 87

PAGE 99

88 [15] W. W ang, I. F. Akyildiz, \A New Signaling Proto col for In tersystem Roaming in NextGeneration Wireless Systems," IEEE Journal on Sele cte d A r e as in Comm. v ol.19, no.10, pp.2040-2052, Oct. 2001. [16] C. P erkins, \IP Mobilit y Supp ort," Internet RF C 2002 Octob er 1996. W ork in progress. [17] S. Mohan and R. Jain, \Tw o User Lo cation Strategies for P ersonal Comm unications Services," IEEE Personal Communic ations v ol.1, no.1, pp.42-50, 1994. [18] I. F. Akyildiz and J. S. M. Ho, \On Lo cation Managemen t for P ersonal Comm unications Net w orks," IEEE Communic ations Magazine v ol.34, no.9, pp.138-145, Septemb er, 1996. [19] J. S. M. Ho, I. F. Akyildiz, \Dynamic Hierarc hical Database Arc hitecture for Lo cation Managemen t in PCS Net w orks," IEEE/A CM T r ans. Networking v ol.5, no.5, pp. 646661, Oct. 1997. [20] R. Jain, Y.B. Lin, C. Lo, and S. Mohan \A Cac hing Strategy to Reduce Net w ork Impacts of PCS," IEEE Journal on Sele cte d A r e as in Comm. v ol.12, no.8, pp.14341444, Oct. 1994. [21] N. Shiv akumar and J. Widom, \A New Signaling Proto col for In tersystem Roaming in Next-Generation Wireless Systems," IEEE Journal on Sele cte d A r e as in Comm. v ol.19, no.10, pp.2040-2052, Oct. 2001. [22] J. Ho and F. Akyildiz, \Lo cal Anc hor Sc heme for Reducing Signaling Costs in P ersonal Comm unications Net w orks," IEEE/A CM T r ans. on Networking v ol.4, no.5, pp.709725, Octob er 1996. [23] R. Jain and Y.B. Lin, \An Auxiliary User Lo cation Strategy Emplo ying F orw arding P oin ters to Reduce Net w ork Impacts of PCS," Wir eless Networks 1 pp.197-210, 1995. [24] J. Z. W ang, \A F ully Distribute Lo cation Registration Strategy for Univ ersal P ersonal Comm unication Systems," IEEE Journal on Sele cte d A r e as in Comm. v ol.11, no.6 pp.850-860, Aug. 1993. [25] B. R. Badrinath, T. Imielinski and A. Virmani, \Lo cating Strategies for P ersonal Comm unication Net w ork," Pr o c. Workshop Networking of Personal Communic ations Applic ations Dec. 1992. [26] V. Anan tharam, M. Honig, U. Madho w, and V. W ei, \Optimization of a Database Hierarc h y for Mobilit y T rac king in a P ersonal Comm unications Net w ork," Performanc e Evaluation V ol. 20, pp. 287-300, 1994. [27] H. W. Hw ang, M. F. Chang and C. C. Tseng, \A Direction-Based Lo cation Up date Sc heme with A Line-P aging Strategy for PCS Net w orks," IEEE Communic ations L etters v ol.4, no.5, Ma y 2000. [28] J. Li, H. Kameda and K. Li, \Optimal Dynamic Mobilit y Managemen t for PCS Netw orks," IEEE/A CM T r ans. on Networking v ol.8, no.3, pp.319-327, June, 2000.

PAGE 100

89 [29] G. Krishnam urthi, M. Azizoglu and A. K. Somani, \Optimal Distributed Lo cation Managemen t in Mobile Net w orks," Mobile Networks and Applic ations v ol.6, pp.117124, 2001. [30] I. F. Akyildiz and J. S. M. Ho, \Dynamic Mobile User Lo cation Up date for Wireless PCS Net w orks," A CM-Baltzer J. Wir eless Networks v ol.1, no.2, pp.187-196, July 1995. [31] H. Xie, S. T abbane and D. Go o dman, \Dynamic Lo cation Area Managemen t and P erformance Analysis," Pr o c. IEEE VTC'93 pp. 536-539, Secausus, NJ, USA. [32] A. Bar-No y I. Kessler and M. Sidi, \Mobile User: T o Up date or Not to Up date," A CM-Baltzer J. Wir eless Networks v ol.1, no.2, pp.175-185, July 1995. [33] I. F. Akyildiz, J. S. M. Ho and Y. B. Lin, \Mo v emen t-based Lo cation Up date and Selectiv e P aging for PCS Net w orks," IEEE/A CM T r ans. Networking v ol.4, no.4, pp.629636, Aug. 1996. [34] V. W. S. W ong and V. C. M. Leung, \An Adaptiv e Distance-Based Lo cation Up date Algorithm for Next-Generation PCS Net w orks," IEEE Journal on Sele cte d A r e as in Comm. v ol.19, no.10, pp.1942-1952, Oct. 2001. [35] C. Rose, \Minimizing the Av erage Cost of P aging and Registration: A Timer-based Metho d," A VM-Baltzer J. Wir eless Networks v ol.2, no.2, pp.109-116, 1996. [36] L. P Araujo and J. R. Boisson de Marca, \P aging and Lo cation Up date Algorirhms for Cellular Systems," IEEE T r an. on vehicular te chonolo gy v ol.49, no.5, pp.1606-1614, 2000. [37] C. Rose and R. Y ates, \Minimizing the Av erage Cost of P aging Under Dela y Constrain ts," A CM-Baltzer J. Wir eless Networks. v ol.1, no.2, pp.490-495, July 1995. [38] U. Madho w, M. L. Hoing and K. Steiglitz, \Optimization of Wireless Resource for P ersonal Comm unications Mobilit y T rac king," IEEE INF OCOM'94 pp.577-584. [39] P J. McCann and I. Hiller, \An In ternet Infrastructure for Cellular CDMA Net w orks Using Mobile IP ," IEEE Personal Communic ations v ol.7, no.4, pp.26-32, Aug. 2000. [40] Ramac handran Ramjee, Kannan V aradhan, Luca Salgarelli, Sandra R. Th uel, ShieY uan W ang and Thomas La P orta, \HA W AI I: A Domain-Based Appro cah for Supp orting Mb oilit y in Wide-Area Wireless Net w orks," IEEE/A CM T r ans. of Networking v ol.10, no.3, pp.396-410, June 2002. [41] A. T. Campb ell, and J. Comez, \An Ov erview of Cellular IP ," IEEE WCNC'99 New Orleans, pp. 606-610, Sept. 1999. [42] Ev a Gustafsson, Annik a Jonsson and Charles P erkins \Mobile IP Regional Registration," Internet dr aft, dr aft-ietf-mobileip-r e g-tunnel-04.txt w ork in progress, Marc h 2001. [43] Jiang Xie and Ian F. Akyildiz, \A Distributed Dynamic Regional Lo caion Managemen t Sc heme for Mobile IP ," Pr o c. IEEE INF OCOM 2002 pp. 1069-1078, 2002.

PAGE 101

90 [44] I. R. Chen, T. M. Chen and C. Lee, \Agen t-Based F orw arding Strategies for Reducing Lo cation Managemen t Cost in Mobile Net w orks," Mobile Networks and Applic ations v ol.6, pp.105-115, 2001. [45] W. Ma and Y. F ang, \Tw o-Lev el P oin ter F orw arding Strategy for Lo cation Managemen t in PCS Net w orks," IEEE T r an. on Mobile Computing v ol.1, no.1, pp.33-45, Jan uary-Marc h, 2002. [46] W enc hao Ma and Y uguang F ang, \An P oin ter F orw arding Based Lo cal Anc horing Sc heme (POFLA) for Wireless Net w orks," tec hnical rep ort 2003. [47] Y. B. Lin and S. K. DeV ries, \PCS Net w ork Signaling Using SS7," IEEE Communic ation Magazine v ol.33, pp.44-55, June, 1995. [48] Y. F ang, I. Chlam tac and Y.B. Lin, \P ortable Mo v emen t Mo deling for PCS Net w orks," IEEE T r ans. V eh.T e chnol. v ol.87, no.8, pp.1347-1384, August 1999. [49] Y. F ang, \Mo v emen t-based Lo cation Managemen t and T radeo Analysis for Wireless Mobile Net w orks," Accepted for publication in IEEE T r ansactions on Computers [50] Y. F ang, \General Mo deling and P erformance Analysis for Lo cation Managemen t in Wireless Mobile Net w orks," IEEE T r ansactions on Computers 51 (10), pp.1169-1181, Octob er 2002. [51] Y.B. Lin and Imric h Chlam tac, \Wireless and Mobile Net w ork Arc hitectures," John Wiley & Sons, Inc, 2001. [52] S. T abbane, \An Alternativ e Strategy for Lo cation T rac king," JSA C v ol.13, no.5, pp.880-892, June, 1995 [53] G. P P ollini and Chih-Lin I, \A Prole-Based Lo cation Strategy and Its P erformance," IEEE JSA C v ol.15, no.8, pp.1415-1424, Octob er, 1997. [54] A. Aljadhai and T. F. Znati, \Predictiv e Mobilit y Supp ort for Qos Pro visioning in Mobile Wireless En vironmen ts," IEEE JSA C v ol.19, no.10, pp.1915-1930, Octob er 2001. [55] S. Choi and K. G. Shin, \Exploiting P ath/Lo cation Information for Connection Admission Con trol in Cellular Net w orks," IEEE WCNC'2000 Chicago, Sept. 2000. [56] B. Liang and Z. J. Hass, \Predictiv e Distance-Based Mobilit y Managemen t for PCS Net w orks," IEEE INF OCOM'99. New Y ork, Marc h, 1999. [57] T. Liu, P Bahl and I. Chlam tac, \Mobilit y Mo deling, Lo cation T rac king, and T rajectory Prediction in Wireless A TM Net w orks," IEEE JSA C v ol.16, no.6, pp.922-936, August, 1998. [58] Y. B. Lin, \Mo deling T ec hniques for Large-Scale PCS Net w orks," IEEE Communic ations Magazine v ol.35, no.2, pp.102-107, F eb. 1997. [59] A. Bhattac hary a, S. D. Das, \LeZi-Up date: An Information-Theoretic Approac h to T rac k Mobile User in PCS Net w orks," IEEE MOBICOM'99 Seattle W ashington, 1999. [60] Y. B. Lin, \Ov erro w Con trol for Cellular Mobilit y Database," IEEE T r ans. on V ehicular T e ch. v ol.49, no.2, pp.520-530, Marc h, 2000.

PAGE 102

91 [61] \Opnet Sim ulation User Reference Do cumen ts," Bethesda, MD, OPNET T ec hnologies, Inc. [62] I. F. Akyildiz, Y. B. Lin, W. R. Lai and R. J. Chen, \A New Random W alk Mo del for PCS Net w orks," IEEE Journal on Sele cte d A r e as in Comm. v ol.8, no.7, pp,1254-1260, July 2000. [63] R. Kalden, I. Meiric k and M. Mey er, \Wireless In ternet Access Based on GPRS," IEEE Personal Communic ations v ol.7, no.1, pp.8-18, April, 2000. [64] F abio M. Chiussi, Denis A. Khotimsky and San tosh Krishnan, \Mobilit y Managemen t in Third-Generation All-IP Net w orks," IEEE Communic ations Magazine pp.124-135, Septem b er, 2002. [65] 3GPP2, \IP Net w ork Arc hitecture Mo del for CDMA2000 Spread Sp ectrum Systems," v er.1.0.0. Oct. 2000. [66] B. Caceres and V. N. P admanabhan, \F ast and Scalable Handos for Wireless In ternet w orks," IEEE MOBICOM'96 New Y ork, pp.56-66, 1996. [67] J. C. S. W u, C. W. Cheng, N. F. Huang and D. K. Ma, \In telligen t Hando for Mobile Wireless In ternet," Mobile Networks and Applic ations v ol.6, pp.67-79, 2001. [68] T. Zhang, E. V. D. Berg, J. Chennik ara, P Agra w al, J. C. Chen and T. Lo dama, \Lo cal Predictiv e Resource Reserv ation for Hando in Multimedia Wireless IP Net w orks," IEEE Journal on Sele cte d A r e as in Comm. v ol.19, no.10, pp.1931-1941, Oct. 2001. [69] J. S. M. Ho and I. F. Akyildiz, \Mobile User Lo cation Up date and P aging under Dela y Constrain ts," A CM-Baltzer Journal of Wir eless Networks (WINET) v ol.1, no.4, pp.413-425, Decem b er, 1995. [70] C. E. P erkins and K. Y. W ang, \Optimized Smo oth Handos in Mbilie IP ," Pr o c. ISCC'99 pp.340-346, Red Sea, EGYPT, 1999. [71] Charles P erkins and Da vid B. Johnson, \Route Optimization in Mobile IP ," dr aft-ietfmobileip-optim-11.txt Septem b er 2001. W ork in progress.

PAGE 103

BIOGRAPHICAL SKETCH W enc hao Ma receiv ed the BS and MS degrees from Beijing Univ ersit y of P osts and T elecomm unications, Beijing, China, in 1995 and 1998, resp ectiv ely F rom 1998 to 1999, he w as an engineer with China T elecom. F rom 1999 to 2000, he w as a Ph.D studen t in the Departmen t of Electrical and Computer Engineering at New Jersey Institute of T ec hnology He is curren tly pursuing the Ph.D. degree in the Departmen t of Electrical and Computer Engineering, Univ ersit y of Florida, Gainesville, where he is a Researc h Assistan t in the Wireless Net w orks Lab oratory (WINET). His researc h in terests include wireless m ultimedia net w orks, mobilit y managemen t, mobile computing and IP mobilit y managemen t. 92


Permanent Link: http://ufdc.ufl.edu/UFE0001252/00001

Material Information

Title: Mobility Management for Wireless Mobile Networks
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0001252:00001

Permanent Link: http://ufdc.ufl.edu/UFE0001252/00001

Material Information

Title: Mobility Management for Wireless Mobile Networks
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0001252:00001


This item has the following downloads:


Full Text










MOBILITY MANAGEMENT
FOR
WIRELESS MOBILE NETWORKS

















By

WENCHAO MA


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


2003




































Copyright 2003

by

Wenchao Ma

















To my wife, Xinglin Liu,

for her support, encouragement

and love














ACKNOWLEDGMENTS


I would like to express my gratitude to all those who have helped me to complete this

dissertation. I am deeply indebted to my advisor, Prof. Yuguang Fang, for his moral support

and encouragement through the years for Ph.D study. I also want to thank my committee

members, Prof. Alan D. George, Prof. Chien-Liang Liu, Prof. Janise McNair and Prof.

Sartaj. K. Sahni, for their interests and comments. Special thanks go to my colleagues in

the Wireless Networks Laboratory (WINET) for their help and support in my research work.

This material is based upon work supported by the National Science Foundation under

Grant No. ANI-0093241. Any opinions, findings, and conclusions or recommendations

expressed in this material are those of the authors and do not necessarily reflect the views

of the National Science Foundation.















TABLE OF CONTENTS


page

ACKNOWLEDGMENTS ............................... iv

LIST OF TABLES ................... .............. vii

LIST OF FIGURES ...................... ............ viii

A BSTR ACT . . . . . . . . .. . x

CHAPTERS

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

1.1 Mobility Management for Wireless Mobile Networks ......... 1
1.2 Mobility Management Schemes in Current Systems .......... 2
1.3 Current Mobility Management Research ....... ...... ... 4
1.4 Organization of the Dissertation .......... .......... 6

2 POINTER FORWARDING BASED MOBILITY MANAGEMENT STRATE-
GIES ..... .............. ................ 7

2.1 Introduction ................... .... 7
2.2 Signaling Network Architecture ......... ........ .. 9
2.3 Basic Mobility Management Procedures of IS-41/GSM MAP .... 10
2.4 Two-Level Pointer Forwarding Scheme . . ..... 11
2.4.1 Mobility Management Procedures . . 11
2.4.2 Cost Functions and Performance Analysis . .... 14
2.4.3 Performance Analysis under Exponential RA Residence Time 19
2.4.4 Sensitivity Analysis to the Residence Time . ... 22
2.5 POFLA Scheme .................. .......... .. 23
2.5.1 The Scheme Overview ................. . .. 23
2.5.2 System Model and Cost Functions . . ..... 25
2.5.3 Performance Evaluations ............... .. 31
2.5.4 Sensitivity to the RA Residence Time . . ... 36
2.6 Discussions .................. ............. .. 37
2.7 Conclusions .................. ............ .. 38

3 USER PROFILE BASED STRATEGY .............. .. .. 39

3.1 Introduction .................. ............ .. 39
3.2 System Description .................. ........ .. 40
3.3 PBS and MPBS Schemes .................. ... .. 41
3.3.1 Profile Based Location Scheme . . ..... 41
3.3.2 Mobility Pattern Based Scheme . . ...... 42
3.4 Cost Evaluation and Simulations .............. .. .. 47
3.5 Numerical Results and Comparison ................. .. 51
v









3.6 Conclusions .................. ......... .. .. 56

4 MOBILITY MANAGEMENT IN MOBILE IP NETWORK ......... 58

4.1 Introduction . . . . . . .. 58
4.2 Related Works ........ ........... ....... 61
4.3 Dynamic Hierarchical Location Management Scheme ........ 63
4.4 Analytical Model .. .... .................... 65
4.5 Numerical Results ........ ................... 70
4.6 Comparison with the IETF Hierarchical Scheme . .... 77
4.7 Improvement of DHMIP Scheme ........... . 79
4.7.1 State Activation ........ ...... ... .. .. 79
4.7.2 Loop Removal ....... ....... ... ..... .. 81
4.8 Conclusions ................. . .. .. 82

5 CONCLUSIONS AND FUTURE WORKS .... . . 84

REFERENCES .............. ........... ... ... 87

BIOGRAPHICAL SKETCH ................. . . 92














LIST OF TABLES

Table page

4.1 The dynamic hierarchical Mobile IP protocol .... . . 65

4.2 The performance analysis parameters (1) ................ . 71

4.3 The performance analysis parameters (2) ................ . 71















LIST OF FIGURES


Figure

2.1

2.2

2.3

2.4

2.5

2.6

2.7


2.8

2.9

2.10


Reference CCS network architecture . ....

The TwoLevelFwdMOVE procedures . ....

The TwoLevelFwdFIND procedures . ....

Relative MOVE and FIND costs of forwarding with 6

Relative MOVE and FIND costs of forwarding with 6

Relative MOVE and FIND costs of forwarding with 6

The effect of variance in residence time (7) with 6
and K2 4 ..... ................

POFLA strategy procedures . ........

The relative costs for the three schemes with P = 0.05

The relative costs for the three schemes with P = 0.05


0.3, K

0.3, K

0.6, K

0.3, K


, G

,G


S 1.5 .

4 . .

S 1.5 .

1.5, K= 4




and p= 1.

and = 3


2.11 The relative costs for the three schemes with P = 0.1, G = 0.1 and 3 = 1.5

2.12 The relative costs for the POFLA and DLA with P = 0.05, G = 0.1 and
3 = 1.5. . .

2.13 The relative net costs for the POFLA and SLA with P = 0.05, G = 0.1 and
3 = 1.5. . .

2.14 The effect of variance of residence time (7) with P = 0.05, G = 0.1 and = 1.5

3.1 Mobility pattern based scheme procedures .. ...............

3.2 The update numbers for three schemes in 24 hours with user residence time
30 m minutes . . . . .

3.3 Simulation network architecture .. .....................

3.4 The location update cost ratio of MPBS scheme to IS-41/C;SI MAP .

3.5 The comparison of the locating time for MPBS and PBS .. .........

3.6 The total costs of MPBS and PBS to IS-41/(;CS MAP with uniform distribution

3.7 The total costs of MPBS and PBS to IS-41/(;CS MAP with linear distribution

3.8 The total costs of MPBS and PBS to IS-41/C(;S MAP with exponential
distribution . . . . . . . . .

viii


page

9

11

12

20

21

22


23

24

32

33









3.9 The effects of user out-of-pattern probability on MPBS and PBS with paging
cost = 0.1, CM R 1 ......... ...... .. . . 55

3.10 The effects of user out-of-pattern probability on MPBS and PBS with paging
cost = 0.5, CM R = 1 ......... ...... .. . . ...... 55

4.1 The MIP location registration and packet routing . . ..... 60

4.2 The DHMIP location registration and packet delivery . . .... 64

4.3 Comparison of the total costs for different schemes . . 72

4.4 Comparison of the total costs for uniformly distributed K under different
parameters .................. ............. ..73

4.5 Comparison of the total costs under different K distributions . ... 74

4.6 Comparison of the total costs under different K . . 74

4.7 Comparison of the total costs under different K distributions with optimal K 75

4.8 The optimal number of hierarchy levels for uniform distribution ...... ..75

4.9 The effect of the variance of the FA residence time () . ..... 76

4.10 The comparison with the IETF hierarchical scheme . . ..... 79

4.11 The total cost for the enhanced DHMIP scheme ............... ..81

4.12 The total cost with loop removal .................. ..... .. 83














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



MOBILITY MANAGEMENT
FOR
WIRELESS MOBILE NETWORKS




By

Wenchao Ma

August 2003

Chair: Yuguang '\!li I. 1 i" Fang
M. i.r Department: Electrical and Computer Engineering

Mobility is an important characteristic of wireless mobile networks. The location of a

called mobile user must be determined within a certain time limit before any service can

be delivered. Thus, mobility management is a key component for the wireless networks to

effectively deliver wireless internet services. In the current wireless systems, the mobility

management is accomplished by a two-tier hierarchical architecture consisting of Home Lo-

cation Register (HLR) and Visitor Location Register (VLR). With the increase of the user

population and service types, the current mobility management scheme faces serious chal-

lenges. Many research works have been carried out to improve the mobility management

procedures for the future generation wireless communication networks. In our research, we

propose a few new mobility management schemes based on either pointer forwarding or user

mobility patterns. In the pointer forwarding based schemes, the user mobility management

operation is improved by two kinds of pointer setup procedures and an additional level of

management entity-Mobility Agent (\1.A). The MAs can be distributed among the network

based on the current traffic load and user mobility so that the new schemes are more adap-

tive and robust while minimizing the total signaling traffic. Motivated by the observation

x









that many mobile users follow some daily routines which can be used to predict the users'

locations, we propose a user mobility pattern based scheme called MPBS. In this scheme,

the users' locations can be determined with less signaling exchange according to the user

profiles. We show that this scheme can significantly reduce signaling traffic. Our third re-

search thrust is the mobility management for Mobile IP networks. Currently, the mobility

over the Internet is addressed by the Mobile IP protocol. In our research, we present a

dynamic hierarchical Mobile IP mobility management strategy (DHMIP). In the DHMIP

scheme, an FA hierarchy is set up dynamically among the network and is specific for every

mobile user. We also develop a rigorous analytical model to evaluate the performance. The

results show that the DHMIP scheme solves the heavy location update traffic problem in

Mobile IP protocol effectively.














CHAPTER 1
INTRODUCTION


1.1 Mobility Management for Wireless Mobile Networks

The wireless mobile communication networks and the Internet have been experiencing

explosive growth since the early 1990s, and it is believed that the user population will be

steadily increasing in the future [1-4]. The wireless network is the only way to achieve ubiq-

uitous communication and computing [5]. The ample and rich information on the Internet

fuels the demands for accessing internet applications through wireless networks. The small

size and high processing capability of the user equipment make this possible.

Unlike the ordinary wired networks, there is not a fixed relationship between a mobile

terminal ID and its location in wireless networks. In the wireless communication networks,

users can move to anywhere in the service area covered by the networks. The terminal IDs

do not implicitly provide the location information of the users. In order to provide the users

with services, there must be some way for systems to locate the users efficiently. This is the

concept of mobility management in wireless communication systems [6]. Generally i ... iii-.

the mobility management includes location registration (update) and call delivery [7,8]. With

the dramatic increase of the user population, the signaling traffic generated by the current

mobility management strategies will consume a lot of system resources (wireless and wired

bandwidth). For the future generation wireless communication systems, the location areas

(LAs) become smaller than before in order to increase the system capacity [9]. Smaller

LA size can help systems utilize radio spectrum more efficiently and reduce the packet

transmission latency more significantly. However, this also makes the mobility management

signaling traffic worse [10].

The Mobile IP is the standard enabling terminal mobility over the Internet [11]. The

IP protocol has been designed for wired networks. There are two major functions for the

terminal IP addresses in the Internet. An IP address is used to identify a particular end

system in the whole network and is also used to find a route between the endpoints. The
1









mobility-enabling protocol for the Internet, Mobile IP, extends the IP protocol by allowing

two IP addresses associated with one mobile host. A mobile host is assigned a permanent

address in its home network and can obtain a temporary one in the visited network. In order

to guarantee continuing services, the mobile host is required to register with its home agent

to set up a binding between the two IP addresses. However, Mobile IP is not a good solution

for users with high mobility [12, 13]. The location update and registration messages can

result in heavy traffic burden to the Internet. Moreover, if a user is far away from his/her

home network or the home agent processing capability is overwhelmed by the huge volume

of update messages, the signaling delay for the location update could be intolerable, leading

to the possible loss of a large amount of in-flight packets.

In this dissertation, we focus on mobility management research for wireless mobile

networks and attempt to solve the aforementioned problems in more efficient ways. In what

follows, we use the location area (LA) and registration area (RA) interchangeably because

they are the two terms used in the IS-41 and GSM MAP protocols for the same concept.
1.2 Mobility Management Schemes in Current Systems

In wireless cellular networks, a user's current location must be determined first before

any service can be delivered to the corresponding wireless access point. In such networks,

the service area is divided into cells. Each cell is primarily served by one base station,

although a base station may serve one or more cells [14]. Base stations are the users'

radio ports. An RA consists of an ..-.-regation of a number of cells, forming a contiguous

geographical region. The location management protocols consist of two major parts: location

registration (or location update) and call delivery. In the location registration procedure, the

mobile terminal updates its current location information to some network databases, and the

information can be retrieved for the future call delivery procedure. Currently, both the IS-

41 and C(;\! MAP share the same characteristics and both use a two-tier system consisting

of Home Location Register (HLR) and Visitor Location Register (VLR) databases. The

signaling network used to set up calls is distinct from the network used to actually transport

the information contents of the calls. It is the signaling network, connected to the databases,

that accomplishes the location management task for the wireless communication network. In

real systems, a VLR can be in charge of one or multiple RAs depending on the user density.









Without loss of generality, we assume that there is one VLR for every RA in our study. A

VLR stores the profiles of the users who are currently residing in its charge area. The HLR

stores the users' profiles and the IDs of current serving VLRs in addition to some entries such

as security and billing information. According to the current mobility management strategy,

a mobile terminal keeps monitoring the broadcast signal containing the ID information of

cells and RAs If the mobile terminal finds that it changes RA, it will send a location

update message to the new VLR via the current serving base station. Upon receiving the

update message, the new VLR will forward it to the user's HLR to request the user's profile

and finish the AAA procedures at the same time. The HLR will update the user's serving

VLR ID in its database and send a de-registration message to the user's old VLR. The old

VLR deletes the user's entry in its database and responds with an acknowledgment message

to the HLR. In the call delivery procedure, when a call for this user is initiated by some

other caller, the location request message is sent to the user's HLR to find out the user's

current location. The HLR forwards the message to the user's current serving VLR. The

current VLR can then determine the user's current serving cell by paging all the cells in its

charging area. After finding out the user's current access point (the BS), the VLR sends

a responding message back to the HLR with a temporary number allocated to the user for

routing purpose. Then, a traffic channel can be set up between the caller and the callee after

the HLR receives the number and forwards it to the caller's serving switch. It is obvious

that, in this scheme, the location update traffic will increase dramatically with the increase

of the user population and the reduction of the RA size. The HLR may become a bottleneck

for the wireless networks [15].

The differences of the network organization between wireless communication network

and the Internet introduce differences in the way the user mobility is dealt with. The user

mobility over the Internet is accomplished through the Mobile IP protocol [16]. In the Mobile

IP protocol, a mobile host has a permanent address (home address) registered in its home

network and this IP address remains unchanged when the user moves from subnet to subnet.

A home agent is a router in a mobile host's home network, which can intercept and tunnel

the packets for the mobile host and also maintains the current location information for the

mobile host. If a mobile host roams to a foreign network, the mobile host can obtain a new









IP address from the foreign agent. The new address is the mobile host's care-of address

(CoA) used for packet routing purpose. The CoA for the mobile host will change from

subnet to subnet. In order to maintain continuous services while the user is on the move,

Mobile IP requires the mobile hosts to update their locations to the home agents whenever

they move to a different subnet so that the home agents can intercept the packets and tunnel

them to the user's current points of attachment. Thus, the Mobile IP can achieve continuous

internet access services for mobile users and does provide a simple and scalable solution to

user mobility.
1.3 Current Mobility Management Research

The current research activities in wireless communication networks generally fall into

two categories based on the system database architecture [17,18]. In the first category, the

mobility management strategies are developed, which aim at improving IS-41/C;S\ MAP

scheme while keeping the basic database network architecture unchanged. The advantage

of this type of solution is that it is easily adopted by the current wireless communication

networks without major modification. These schemes are based on centralized database ar-

chitectures inherited from the IS-41/;S\!I MAP standard. The following schemes belong to

this category: .7;l,.in;, hierarchical database architecture [19]; per-user location caching [20];

user PI-.'P1 replication [21]; local .,i,.,-'i, ;. [22] and pointer forwarding [23]. Another cat-

egory of research results lie in completely new database architectures that require a new

set of schemes for location registration and call delivery. The major schemes belonging to

this category are: fully distributed registration scheme [24]; ., 1./1.: '.:.':./ [25] and database

hierarchy [26]. All the above schemes can be considered as static because the mobile termi-

nal is mandatorily required to register its current location to some database every time it

enters a new LA. Some schemes are proposed for the mobile terminal to perform location

registration dynamically [27-29]; in other words, the mobile terminal sends a location up-

date message to the system according to the user's current movement status [30]. Usually,

the dynamic schemes can achieve better performance than static ones; however they are

more complicated in implementation too. Several important dynamic mobility management

schemes are .;l,,in,,- LA monaingmrrint [31]; time-based, movement-based and distance-based

.iini,-i update schemes [32-34]. In our work, the proposed two-level pointer forwarding









and POFLA strategies belong to the static scheme category and the user ,,ii.1.:I.:/1i pattern

based scheme can be considered as dynamic. In general, there is a tradeoff between the

paging cost and the connection setup delay [35, 36]. So, there are some research works in

this area attempting to minimize the paging cost under a given delay constraints, such as

paging under .. 1l,,i constraints [37] and update and paging under .. 1,,,i constraints [38].

The IP network administration is based on the network administrative regions. So the

IP mobility can be broadly classified into macro-mobility and micro-mobility. The macro-

mobility is for the case that an MH roams across different administrative domains. The

macro-mobility occurs less frequently and usually involves larger timescales [12]. The Mobile

IP can ensure that the mobile users reestablish communication connections after a move

during macro-mobility. The micro-mobility means the MH movement across multiple subnets

within a single network domain. For micro-mobility, which occurs quite often, the Mobile IP

paradigm needs to be enhanced. Most of the related works attempt to improve the Mobile IP

micro-mobility handling capability [13]. The micro-mobility proposals can be classified into

two distinct categories [39]. The first is the routing update approaches, in which IP routing is

used to direct traffic toward the MHs so that the multiple reencapsulation and decapsulation

along the FAs are avoided. This category includes Hoi:JfT,-Aware Wireless Access Internet

Infrastructure (HAWAII), TeleMIP and Cellular IP. The HAWAII is a separate routing

protocol to handle micro-mobility [40]. The scheme hinges on the assumption that most

user mobility is local to an administrative domain of the networks. An MH entering a new

foreign network is assigned a new CoA and retains its CoA unchanged while moving within

the foreign domain. For macro-mobility, the HAWAII uses the traditional Mobile IP. In this

sense, the scheme can be considered as an enhanced Mobile IP. The TeleMIP [12] scheme tries

to localize the scope of location update messages by introducing a new management entity.

In the TeleMIP, a mobility agent is in charge of one region, handling the CoA addresses for

those MHs roaming in the region. TeleMIP can enhance the performance in the IP support

over the cellular networks. The Cellular IP scheme [41] is another strategy supporting local

mobility in cellular environment. In this scheme, the network is connected to the Internet

through gateway routers and the roaming between gateways is managed by Mobile IP while

the mobility within access networks is handled by a new designed protocol called cellular









IP. Another category of the micro-mobility schemes is the hierarchical tunnel approaches,

characterized by their reliance on a tree-like structure of FAs. The regional registration

protocol belongs to this category. The Mobile IP regional registration protocol [42] is an

extension of the Mobile IP scheme. The new protocol employs the FA hierarchy to localize

the registration traffic. In the regional registration scheme, the network architecture is

centralized. The distributed .7lin,.', regional location inainio -nriit scheme [43] is a dynamic

hierarchy strategy and the regional network size is adjusted based on the user's current traffic

load and mobility information. This strategy can be considered as an extension of the IETF

regional registration scheme to make it more flexible and adaptive. In our research, we

propose a new location management scheme for Mobile IP network- ., in;i.-, hierarchical

Mobile IP (DHMIP). The new scheme architecture is dynamic and the FA hierarchy is

varying based on the user behaviors.
1.4 Organization of the Dissertation

This dissertation is divided into five chapters. Following this chapter, chapter 2 describes

the two-level pointer forwarding and POFLA strategies. In this chapter, we analyze the

performance of the new schemes and compare them with the IS-41/GSM MAP scheme

theoretically. We also compare the performance of the two new schemes with those of the

local anchor and per-user forwarding schemes in this chapter. At the end of chapter 2, we

briefly discuss the relationship between our new schemes and those previously mentioned and

give some alternative ways for implementation. In chapter 3, a new mobility management

scheme, MPBS, for next generation wireless communication system is proposed. In this

scheme, the mobile terminal will perform location update based on the user mobility pattern.

In this chapter, we describe the details of the PBS and the new MPBS schemes and the

performance evaluations are carried out. Mobile IP is the protocol achieving mobility over

the IP network. However, the micro-mobility in Mobile IP protocol is not efficient for most

applications. Therefore we introduce a new location management scheme for Mobile IP

network in chapter 4 and an analytical model is developed in that part to evaluate the

performance. In the last chapter, we present the conclusions and future research work in

this area.














CHAPTER 2
POINTER FORWARDING BASED MOBILITY MANAGEMENT STRATEGIES


2.1 Introduction

In this chapter, the two new location management strategies for wireless communication

networks based on pointer forwarding technology will be presented.

According to the IS-41/GSM MAP strategies, a mobile terminal performs location up-

date (registration) to the HLR every time the user crosses a Registration Area (RA) bound-

ary, and deregisters at the previous VLR [7,8]. These location management procedures will

incur high signaling traffic. If many users are far away from their HLRs, heavy signaling

traffic over the network can occur. This problem becomes worse with the increase of the

number of mobile users. The local anchor scheme reduces the signaling traffic by choosing

a local anchor for each user [22]. In this scheme, a VLR close to a user is selected as the

local anchor for him or her. Whenever the user moves from one RA to another, the mobile

terminal will perform location update to the local anchor. The local anchor for a mobile user

will not change unless a call request arrives; at the same time, the HLR is also updated via

the call delivery procedures. When a call request terminating at this user is received by the

HLR, the user can be traced by the local anchor. The local anchor scheme avoids update to

HLR completely at the expense of increasing the local signaling traffic. Some similar scheme

was proposed in [44]. The drawback of this kinds of schemes is that when the user keeps

moving constantly without receiving any call, the updates to local anchor may become costly

too, a similar bottleneck as the HLR is. For example, at the end of conferences or games,

many people move away from one site without receiving calls, and the local anchor for these

people can become a bottleneck. Jain and Lin proposed another scheme called per-user

pointer forwarding scheme [23]. In this scheme, some updates to the HLR can be replaced

by setting up forwarding pointers from the previous VLRs to the new VLRs. When a call

request to a mobile user arrives, the wireless communication network first queries the user's

HLR to determine the VLR which the user was visiting at the previous location update,
7









then follows a chain of forwarding pointers to the user's current VLR to locate the mobile

user. The traffic to the HLR is decreased by the pointer chain; the p. in.11 is the time delay

for tracing the user when a call to the user arrives. The longer the pointer chain, the less the

signaling traffic and the longer the call setup delay. To make sure the setup delay is under

some constraint, a threshold of the pointer chain length is set. The user needs to perform

registration to the HLR whenever the chain threshold is reached.

In order to overcome the drawbacks of the above two schemes, we propose two kind of

location management schemes based on the pointer forwarding technique. One is two-level

pointer fr ,,,i 'iii.,, strategy [45]; another is POFLA scheme [46]. In both of the two schemes,

two kinds of pointers are used. Some VLRs are selected as the Mobility Agents (1\ As), which

will be responsible for the location management in a larger area comparing with the RAs

and can be geographically distributed. The MAs are also involved with longer or higher level

pointers which can minimize the calling setup delay. The shorter or low level pointers chains

are set up between the .,li.I1:ent VLRs. Both schemes can avoid the possible costly updates

to HLR and the traffic congestion in local anchor. More importantly, the thresholds for the

new schemes are two parameters which can provide the flexibility in design comparing with

the one-parameter traditional pointer forwarding strategy. The two schemes are detailed in

the following sections separately.

This rest of this chapter is organized as follows. In section 2.2, we describe the ba-

sic signaling network architecture to facilitate the presentation and analysis of the basic

IS-41/C;S\ MAP protocols, the new two-level pointer forwarding and POFLA strategies.

Section 2.3 introduces the basic IS-41/CS\I MAP location management strategy in detail.

We introduce and analyze the performance of the two-level pointer forwarding scheme in

section 2.4; we also analyze how the user RA residence time can affect the two-level for-

warding scheme performance in this section. Another new pointer forwarding based location

management scheme, POFLA, is introduced and analyzed in section 2.5. In this section, we

also compare the the performances of the new schemes and some other proposed strategies

under various conditions. We briefly discusses the relationships between our new schemes

and some improved ones in section 2.6, and section 2.7 gives the conclusions.










SCP/HLR

Remote A link

RSTP

D link


S/ LSTP
Local A link

SSP
SVLR/
MSC



Registration Area



Figure 2.1: Reference CCS network architecture


2.2 Signaling Network Architecture

The network used to accomplish signaling exchange is distinct from the network used to

actually transport the information contents of the calls. Especially, we assume a Common

Channel Signaling (CCS) network is used to set up calls which use the Signaling System No.7

(SS7) protocols [47]. Fig. 2.1 shows a typical No.7 signaling network architecture. In a CCS

No.7 signaling network, all the base stations in an RA are connected via wire-line network

to an end-office switch, or Service Switching Point (SSP). Each SSP serves an RA. All the

SSPs of different RAs are in turn connected to a higher hierarchical Local Signaling Transfer

Points (LSTP), which are connected to a Regional STP (RSTP). An RSTP connects to all

the LSTPs in one region. In practice, each STP actually consists of two STPs in a mated-

pair configuration for reliability. For the simplicity of presentation, Fig. 2.1 only shows one

of each pair. The RSTPs are also connected to a Service Control Point (SCP). Each SCP is

equipped with an HLR database.

For simplicity, we assume that each VLR is associated with one Mobile Switching Center

(\!SC), which connects the BSs and backbone communication infrastructure (such as Public

Switched Telephone Network (PSTN)). Therefore, we assume that an MSC, an SSP and

a VLR database are associated together to serve an RA. The configuration may vary in

practice; however, the assumption is used only for performance analysis. Since we do not






10

deal with the content of the messages, we assume that the message sizes are equal for all

signaling transactions.
2.3 Basic Mobility Management Procedures of IS-41/C;S\I MAP

In the following sections, we will compare the performances of the new schemes with

that of the basic IS-41/C(;S\ MAP scheme. The new schemes can minimize the overall

signaling traffic burden by revising the current protocols. So we need to define the basic

scheme first. To facilitate the presentation, the following two operations are defined:

MOVE : the user movement from one RA to another;

FIND : determination of the RA where the user is currently in.

We call the MOVE and FIND operations used in current mobility management stan-

dards such as IS-41 or (;Si\ MAP the BasicMOVE and BasicFIND. We present the pro-

cedures in the following pseudocode. We remark that the BasicMOVE and BasicFIND

procedures we present here are simplifications of those in the standards; however, such sim-

plifications do capture the major interactions between the HLR and VLR databases relevant

to our comparative study.

BasicMOVE(

{The mobile terminal detects that it is in a new registration area;

The mobile terminal sends a registration message to the new VLR;

The new VLR sends a registration message to the user's HLR;

The HLR sends a registration cancellation message to the old VLR;

The old VLR sends a cancellation confirmation message to the HLR;

The HLR sends a registration confirmation message to the new VLR;}

BasicFIND (

{Call to a user is detected at the local switch;

if the called / ,,/ Ii is in the same RA, then return;

Switch queries the called r"h I/ 's HLR;

HLR queries the called r, Ili's current VLR V;

VLR V returns the called r/,, li,'s location to HLR;

HLR returns the location to the i.l//l,,i; ";lI1.:}







11


HLR

REGNOT
regnot
REGPTR L1 REGPTR L1 REGPTR L2 regno REGPTR L1 REGPTR L2
SSP/
MSC/
VLR
regptr regptr 11 regptr 12 regptr 12 1 egptr 1_








i=0 i=1 i=2 i=K -1 i=0 i=1 i=1
j=0 j=K2 j=K2 j=K2-1 j=0 j=K2 j=

Figure 2.2: The TwoLevelFwdMOVE procedures


2.4 Two-Level Pointer Forwarding Scheme

2.4.1 Mobility Management Procedures

The two-level pointer forwarding scheme modifies the basic procedures as follows. When

a user moves from one RA to another, it informs the switch (and the VLR) at the new RA

about the old RA. It also informs the new RA about the previous MA it was registered. The

switch at the new RA determines whether to invoke the BasicMOVE or the TwoLevelFwd-

MOVE strategy.

In TwoLevelFwdMOVE, the new VLR exchanges messages with the old VLR or the old

MA to set up a forwarding pointer from an old VLR to the new VLR. If a pointer is set up

from the previous MA, the new VLR is selected as the current MA. The TwoLevelFwdMOVE

procedures do not involve the user's HLR. Fig. 2.2 shows the Two-Level Forward MOVE

procedures with level_l pointers chain threshold limited to 3. Assume that a user moves

from RAa to RAg (these RAs are not necessary to be .ili.,:ent) and RAa is the user's MA.

When the user leaves RAa but before he enters RAb, the user informs the new VLRs and

the level_2 pointers are built from the old VLR to the new VLR. When the user enters RAb,

the chain threshold for level_2 pointer is reached, so RAb is selected as the user's new MA

and a level_l pointer is set up from the old MA to the new MA. At the same time, the

level_2 pointer chain is reset. The similar procedures are used at RAe. A level_l pointer










SCP/HLR
(4) (1) (3)


RSTP




LSTP



VLR
L1 forwarding pointers L1 forwarding pointers L2 forwarding pointers
(2)
incoming call


Figure 2.3: The TwoLevelFwdFIND procedures


is set up from RAb to RAE, and the VLR in RA, is the user's new MA. As the user keeps

moving, in RA,, the threshold for level_2 pointer chain is reached again, while this time the

threshold of the level_l pointer chain is reached too. Instead of exchanging information with

the previous MA, the BasicMOVEO is invoked. The HLR is updated with the user current

location. The messages REGPTR_L1 and REGPTR_L2 are messages from the new VLR to

the old VLR specifying that a level_l or level_2 forwarding pointer is to be set up; messages

regptr_11 and regptr_12 are the confirmations from the old VLR (or MA). In this figure, the

VLRs in RAa, RAb, RAe and RAf are selected as the user's MAs.

The TwolevelFwdFIND procedures are invoked for the subsequent calls to the user from

some other switches. The user's HLR is queried first as in the basic strategy, and a pointer

to the user's potentially outdated MA is obtained. The pointer chain is followed to find out

the user's current location (see Fig. 2.3). As we can see, in the two-level pointer forwarding

scheme, the chain length can be longer than that in the basic pointer forwarding scheme

without increasing the Find penalty significantly. The previous study [23] shows that more

saving can be obtained with a longer chain. However, the pointer chain length is limited

by the delay restriction requirement. By appropriately tuning the two thresholds in our

schemes, we can mitigate the signaling cost without too much increase of call setup delay.

The two-level pointer forwarding procedures can be described by the following pseu-

docode (we use the shared global variables i and j in the pseudocode).








TwoLevelFwdMO VE()

{/*Initially, i,j are 0*/

if(j < K2 and i < K1)

{The user registers at the new RA/VLR,

,I,. .:'i .the ID of the former RA/VLR and MA;

The new VLR deregisters the user at old VLR;

The old VLR sends ACK and the user's service "r, /l. to the new VLR;

j:= + 1;}

else if(j >= K2 and i < K1)

{ The user registers at the new RA/VLR(MA),

,Ir .:',i the ID of the former RA/VLR and MA;

The new VLR deregisters the user at the old MA/VLR;

The old MA/VLR sends ACK and the user's

service ","/;I. to the new MA/VLR;

i:= i ; := 0;}

else

{ BasicMOVE();

i := 0; j := 0;}}

TwoLevelFwdFIND (

{ A call to the PCS user is detected at a local switch;

if (the called /,, I is in the same RA) return;

The local switch queries the called h"" Ii's HLR;

HLR queries Vo/l. I1;

While(Queried VLR is not the called h", Ii's current VLR);

VLR queries the next VLR in the pointer chain;

/*Now the called ,, Ilhi's actual VLR has been found*/

i := 0; := 0;

The called f'I,, i's current VLR sends the user location to HLR;

HLR sends the user location to I.i//,,,l; ,'l'i's switch;}









2.4.2 Cost Functions and Performance Analysis

In this section, we develop an analytic model and study the performance of the two-level

pointer forwarding strategy based on different parameters for different classes of users.

We characterize the classes of users according to their call-to-mobility ratio (CMR).

The CMR of a user is defined as the expected number of calls to a user during the period

that the user visits an RA (notice that the CMR is defined here in terms of the calls received

by a particular user, not calls originated from the user). If calls are received by the user at

a mean rate A and the time the user resides in a given RA has a mean /pi, then the CMR,

denoted as p, is given by


S= A/p. (2.1)

In order to make comparison of the costs, we need to analyze the basic procedures first.

Assume that a user crosses multiple RAs between two consecutive calls. If the basic user

location strategy is used, the user's HLR is updated every time the user moves to a new RA.

If the two-level pointer forwarding strategy is used, the HLR is updated only every K1 K2

moves (K1 and K2 are the level and level_2 pointer chain length threshold respectively),

while forwarding pointers are set up for all the other moves.

We define CB and CF as the total costs for maintaining the location information (loca-

tion updating) and locating the user (location tracing) between two consecutive calls for the

basic strategy and the two-level forwarding strategy, respectively. The following notations

will be used in our analysis.

m : the cost of a single invocation of BasicMOVE;

M : the total cost of all the BasicMO VESs between two consecutive calls;

F : the cost of a single BasicFIND;

M' : the expected cost of all TwoLevelFwdMOVEs between two consecutive calls;

F' : the average cost of the TwoLevelFwdFIND;

S1 : the cost of setting up a forwarding pointer levell) between MAs during a
TwoLevelFwdMOVE;

S2 : the cost of setting up a forwarding pointer (level_2) between VLRs during a
TwoLevelFwdMO VE;








TI : the cost of traversing a forwarding pointer (level_l) between MAs during a
TwoLevelFwdFIND;
T2 : the cost of traversing a forwarding pointer (level_2) between VLRs during a
TwoLevelFwdFIND;
K1 : the threshold of level_l pointer chain;
K2 : the threshold of level_2 pointer chain;
a(i) : the probability that there are i RA crossings between two consecutive calls.
Then, we have

CB =M F = /p + F. (2.2)

C = M'+ F'. (2.3)

Now, we can derive formulas for M' and F' as follows. Suppose a user crosses i RA
boundaries between two consecutive calls. The HLR is updated L[ 2] times. There are
also [iJ [> level pointer creations (every K2 moves may require a level_- pointer
creation but sometimes the HLR is updated and level pointer is not set up). The level_2
pointers are created for all the rest i L- moves. Thus, we obtain


M{L =m + L-S + ( -S2} (i).

(2.4)

The cost of F' is derived as follows. After the last BasicMove operations (if any), the user
i- I Ki K2 | i- Ki K2
traverses I-LK2 level-1 pointers and i- j KK2- LK2 K2 level_2
pointers. Thus, we obtain

F' o i K1K2T K1K2


K2
LK J1K2JT2a(i). (2.5)


In order to evaluate a(i), we make the following assumptions.
1. The call arrivals to a user form a Poisson process with arrival rate A.







16

2. The residence time of a user at a registration area is a random variable with a general

density function fm(t) and the Laplace transform is

ro
f*(S) = fm(t)e- dt.
Jt=0

The expected residence time of a user in an RA is 1/tp. We denote g = f*(A) for convenience.

With these assumptions, (2.4) and (2.5) can be derived as follows.


M/ = M + S, + i -S2 Ce2}
i00

Sj iS^(i)+j \- (S- S2)a(i)+^ (m- S)a().
i=0 i=0 i=0
x Y Z

X can be simplified from the definition of a(i),
00 S2
X S2 a(i) -.
i=0 P

The probability a(i) can be expressed as (see [48] for the detailed derivation),


a (i)


(1 g)2gi-1


(2.6)


Y = (Si S2i ai).
i=0 2


Let i = jK2 + k, then,


a(jK2 + k)


(1 -g)2 (g)K2 k y- xk.
pg


( g)2
y ,
P9


where


z K2
g -g


x = g.








We can rewrite,


oc K2-1
Y y(S S2) zJxk
j=0 k=0
y(S S2)(1- K2) (jJ
1--x
j=0
yz(S S2)(1- K2)
(1 -x)(1- z)2
(1 g)(SI S2)gK2-1
p(l gK2)


Similarly,


Z ( S1)
i=-


_ l _i a(i)
LK1K2j


(1 g)(m S)gK1K2-1
p(l gKiK2)


So, we obtain


/ S2 (1 g)gK2-1(S1 S2)
M -+ -+
p p(l gK2)

We can derive F' in a similar way,

i-o i K1


g)gK1K2-1 (m
p(l gKiK2)


S)
(2.7)


i KlK2
iK2


-i L K K2K


= F+(T -K2T2) [ I I2 0()
i=0
U

+T2 (i K K1 K2)a(i)

V
Let i jK1K2 + k, then,


a(jKiK2 + k)


( g)2
Ypg
P9


(1 g)2(gKK2gk
P9


z =gKK2 x g.


where


z k
yz x









2T i- i K K2
U (T K2T2) [ KK2)
i=0 2
oo K1K2-1
S(Ti K2T2) a(jiK,2 +k)
j=0 k=0
oo K1K2-CI 111


0o KK2- I 0 k
(TI K2T2) Y l-2 y


j=0 k O
oo K1-1 K2-
S(Ti r^K2T2) y ^ (5 xnK2 m
j=0 n=O m=O
(T K2T2)(- g)[gK2 KgK1K2 + (K 1)g(K+1)K2]
pg(1 gKK2)(1- gK2)



V T2 i L--KK /J1/K2)0(i)
i=0
o0 K K2 --
oo KCK2-1-
T2 ka(jKK2 + k)
j=0 k=0
oo KK2- 1
yT2 z kXk
j-0 k-0
[1 KK2gKIK2-1 + (K1K2 1)gKK2 T2
(l gK1K2)

So we have

ST2[1 KiK2gK1K2-1 (KK2 1)gK1K2]
p(l gKK2)
(TI K2T2)( g)[gK2 KgK1K2 ( )KK ) +1)K2]
S(2.8)
pg(1 gKK2)( gK2)

For demonstration purposes, we assume that the RA residence time of a user is Gamma

distributed with mean 1/p. The reason that Gamma distribution is selected is its flexibility

in setting various parameters and can be used to fit the first two moments of the field data.

The Laplace transform of a Gamma distribution is


f, (s)= ( ) ----)Y
8s + w








thus, we have,


Sf (A) -( (2.9)

In particular, when 7 1, we have an exponential distribution for the RA residence time.
2.4.3 Performance Analysis under Exponential RA Residence Time

We first consider the situation when the RA residence time is exponentially distributed.

By setting 7 1 in (2.9), we have
1
1+p
Then (2.7) and (2.8) can be rewritten as

S2 S1 S2 m S,
M' + + (2.10)
p (+ p)K2 I (I p)KiK2
F' T2 T2K1K2
p (1 + p)KK2
(T K2T2)[(1 + pKK2 KI(1 + p)K2 + K 1]
[( +p)K+p)(2.11)

From (2.10), (2.11) and (2.3), we obtain

T F 2 + 52 S1 2 m S T2K1K2
C, = F-- + +
p (1 + p)K2 1 ( + p)KK2 1
(Ti K2T2)[(1 + p)KiK2 K (1 + p)K2 K1- 1]
S(2.12)
[(1+ p)KiK2 [(l p)K2 (2 )

We notice that updating the HLR and performing a BasicFIND involve the same number

of messages between HLR and VLR databases, so we can choose m = F. Without loss of

generality, we can normalize m = 1. We also assume that the cost of setting up a forwarding

pointer is about twice the cost of traversing it, since twice as many messages are involved,

i.e., we set S1 = 2T1 and S2 2T2. We consider S2 6 with 6 < 1. Since the level_l pointer

is more expensive than level_2 pointer in terms of setup cost, we can assume 5' = KS2 with

K > 1. It is reasonable to assume that S1 < 1 too. We will see later, however, that the

two-level forwarding strategy can also perform well even with S1 > 1. From (2.2), (2.10),

(2.11) and (2.12), we obtain,


CB = + -, (2.13)
p










(a) The MOVE cost M'/M (b) The FIND cost F'/F (c) The net cost CF/CB
042 1 8 085
K1=2;K2=4 K1=2;K2=4
K1=2;K2=6 K1=2;K2=6
K1=4;K2=4 K1=4;K2=4 08-
K1=4;K2=6 17 K1=4;K2=6
04
040 75 -
16
07
038
1 5 //
0 65

036 1 4 06 -

S055 I
13
0 34 \
0340 5 -1
05
12
Y\0 45
032
1 1 K1=2;K2=4
04 K1=2;K2=6
SK1=4;K2=4
SK1=4;K2=6
03 1 035
0 12 0 1 2 0 1
p(CMR) p(CMR) p(CMR)

Figure 2.4: Relative MOVE and FIND costs of forwarding with 6 = 0.3, K = 1.5


M' (K 1)6p (1 K6)p
6 + (2.14)
M (1+ p)K2 1 ( +p)K1K2 )
F' 6 K1K26
F 2p 2[(1 p)KK2 _

6(K K2)[( + p)KK2 K(1 + p)K2 K1 1]
S(2.15)
2[(1 p),K2 ][( +p)K2 1 (2.1



CF pP 36 (K 1)6 1- (K+ KIK2)6
1+ + + 2
CB 2p (+ p)K2 I (I p)KK2

S6(K K2)[( + p)KK2 K p)K2 K1 1]
(2.16)
2[(1 p)K-K2 ][(l+ P)K2 1]

In Fig. 2.4, 2.5 and 2.6, we plot the costs as functions of CMR for various values of

K1, K2, K and 6. Fig. 2.4(a) shows that under certain conditions (6 = 0.3, K = 1.5), the

two-level forwarding scheme can result in 60% 70% reductions in location update cost

comparing with the basic strategy. However, the Fig. 2.4(b) indicates that the FIND cost

of the two-level forwarding scheme is higher than the basic strategy. The reason is that the

call setup messages for the user need traverse the pointer chain to find the user's current

location. However, as we observe in Fig. 2.4(c), the two-level forwarding strategy can result

in 20% 60% reduction of the total cost. If we study the plots carefully, we can observe

that both the relative MOVE an FIND costs are decreasing functions of p (CMR). When p

is small, the user crosses RAs more frequently. The pointers are needed to be set up and a











(a) The MOVE cost M'/M (b) The FIND cost F'/F (c) The net cost CF/B
0 55 09
S K1=2;K2=4 K1=2;K2=4 K1=2;K2=4
K1=2;K2=6 K1=2;K2=6 K1=2;K2=6
S K1=4;K2=4 K1=4;K2=4 085 K1=4;K2=4
K1=4;K2=6 K1=4;K2=6 K1=4;K2=6
05
08
2
0 75
045
07-


r \1 \065-///
0 65 -
04-\
S1 \-15
06

035- 0 55

05

03 1 045
0 1 2 0 1 2 0 1 2
p(CMR) p(CMR) p(CMR)

Figure 2.5: Relative MOVE and FIND costs of forwarding with 6 = 0.3, K = 4


long chain of pointers have to be traversed, leading to the high setup cost. The improvement

of the total cost increases when the p decreases because most MOVEs do not result in HLR

updates but pointer creations. Consider the Fig. 2.4(a) again, we can have more saving in

the MOVEs with longer pointer chain because more updates to HLR can be substituted with

pointer creations. However, long pointer chain increases the FIND penalty at the same time

(see Fig. 2.4(b)). An advantage of the two-level forwarding strategy is that it can have long

pointer chain without increasing the delay pi n.ill significantly because the pointer chain

can be shortened by the level_l pointers between MAs. Under the assumed conditions,

the maximum pointer chain length can increase from 8 to 24 with only 30% FIND penalty

increase.

The Fig. 2.5 shows the relative cost curves when K increases from 1.5 to 4. As we can

see, even in this case, the cost of setting up a level_l pointer exceeds that of updating HLR,

there is only slight increase of the total cost. The MOVE and FIND costs both increase

because the cost of setting up and traversing level pointers chain increases. Since level

pointer is built up only when the level_2 pointer chain threshold is reached and the number

of level_2 pointers is dominant, therefore the two-level forwarding strategy is not sensitive to

the variation of K. The Fig. 2.6 indicates that the level_2 pointer operation cost 6 has more

effect on the system performance. In Fig. 2.6, 6 is increased from 0.3 to 0.6. The MOVE,










(a) The MOVE cost M'/M (b) The FIND cost F'/F (c) The net cost CF/C
069 25 1
K=2;K=4 K,=2;K2=4
068 K1=2;K2=6 K=2;K2=6
K1=4;K2=4 K1=4;K2=4 095 -
K1=4;K2=6 K =4;K =6
0 67 //

066 21

065 085
0 65 //

064 -08 -0

0 63 1 5 \\
S\ 075
0 62
SK,=2;K2=4
\07 K,=2;K2=6
061 I K1=4;K2=4
K1=4;K2=6
06 1 065
0 1 2 0 1 2 0 1 2
p(CMR) p(CMR) p(CMR)

Figure 2.6: Relative MOVE and FIND costs of forwarding with 6 = 0.6, K = 1.5


FIND and the net costs all increase. Finally, we can observe that for small 6, increasing

pointer chain length reduces the cost of two-level forwarding scheme because the pointer

operations are cheaper.

2.4.4 Sensitivity Analysis to the Residence Time

We now investigate the sensitivity of the performance costs and benefits of the two-

level forwarding scheme to the variance of the user's mobility patterns. We assume that

the call arrivals to a user form a Poisson process, and the RA residence time has a Gamma

distribution.

For a Gamma distribution, the variance is V = _-. In other words, a large 7 implies

a small variance. Fig. 2.7 shows the effect of 7 on M'/M, F'/F and CF/CB. In Fig. 2.7,

we observe that the increase of the variance of RA residence time (smaller 7) causes the

increase of M'/M but the decrease of F'/F; the net effect to CF/CB is not significant.

Consider M'/M for 7 < 1 comparing with the case when 7 = 1, for a given p > 0

(see Fig. 2.7(a)), the large variance of RA residence time implies high variation of the RA

boundary crossing patterns. If the user crosses many RAs, a longer pointer chain will be

created. When the chain limit K1 K2 reaches, the HLR will be updated, resulting in the

increase of M'. On the other hand, if fewer boundaries are crossed, only shorter pointer

chains will be set up, the pointer creation/tracing cost will be saved. The net effect is an












(a) The MOVE cost M/M (b) The FIND cost F'/F (c) The net cost C/CB
037 1 7 085
r=100
r=1
r=0 01 08-
036 1 6
075

035 r1 15 07
r=100
r=1
=0 01 065
034- 1 4 -
06-

0 55 -

032- 1 2 05-

0 45-
031 11
04 r=100
r=1
r=0 01
03 1 035
0 1 2 0 1 2 0 1 2
p(CMR) p(CMR) p(CMR)


Figure 2.7: The effect of variance in residence time (7) with 6 = 0.3, K = 1.5, K = 4 and
K2 4


increase in M'. Now, consider the effect of variance in the RA residence time to the F'/F

(See Fig. 2.7(b)). When the variance is high, the number of RA boundaries the user crosses

between two consecutive calls will vary greatly. When the number is small, the FIND cost

is reduced; when the number is large, the pointer chain could be shortened. The net effect

is a significant improvement in F'/F. The net effect of the variance of the residence time on

total cost ratio CF/CB is not significant for low CMR (see Fig. 2.7(c)).

2.5 POFLA Scheme

We will introduce another pointer forwarding based scheme-Pointer Forwarding Based

Local Anchoring Scheme (POFLA) in this section.

2.5.1 The Scheme Overview

In the POFLA scheme, the basic location update and call delivery procedures are mod-

ified to achieve better performance. A VLR serving a user is selected as the MA for that

user and may change during the user's movement. Since the selection of MAs for the users

is based on their locations, hence the signaling traffic can be distributed evenly among the

network. This can avoid the bottleneck effect normally experienced by the HLR.

The basic location update procedure is modified as follows: every time a user enters

a new RA served by a different VLR, the mobile terminal registers to the new VLR and











HLR

MA MA
H Pointer H Pointer

L Pomters LPon L Pomte L Pomter
VLR1 MSC1 VLR2 MSC2 VLR3 MSC3 VLR4 MSC4 VLR5 MSC5 VLR6 MSC6 VLR7 MSC7 VLR8 MSC8
RAI RA2 RA3 RA4 RA5 RA, RA7 RA8



Figure 2.8: POFLA strategy procedures


informs the new VLR about the old VLR and MA. The old VLR and MA may be the same

VLR the user currently visited. The VLR at the new RA determines what to do based on the

information from the mobile location update behaviors. The new VLR has three options: it

can request the old VLR to set up a pointer to itself, which is called Low Level Pointer (or L

Pointer) in our scheme; it can update the MA and request to set up a pointer from the MA

to itself, this is called High Level Pointer (or H Pointer); and it can also decide to update the

user's new location to the HLR directly and it itself becomes the new MA. Fig. 2.8 shows

the location update and call delivery procedures in the POFLA scheme with the H pointer

number limit being three. Assume a mobile user moves from RA1 to RA8 (these RAs are

not necessary to be .,.li.' Ient) and VLR1 is the user's current MA. At the beginning, the user

is in RA1 and VLR1 is the user's current serving VLR. The VLR1 is selected as the user's

current MA because either the user just receives an incoming call in RA1 or the VLR1 just

updates the user's new location to the HLR. When the user leaves RA1, but before enters

RAs, the mobile terminal informs the new VLR and a pointer chain consisting of L pointers

is set up just as in the per-user forwarding scheme [23]. When the user enters RAs, the

chain threshold for L pointers is reached. In this situation, the VLR3 will update the user's

new location to the current MA, i.e. VLR1. At the same time, the L pointer chain is reset.

The same procedure is used in VLR5 and the previous H pointer is reset. If the user keeps

moving, in RA7, the threshold for L pointer chain is reached again. This time, the limit of

the H pointer number is reached too. Instead of exchanging information with the previous

MA and setting up a new H pointer, the VLR7 will update the user's location to the HLR

directly and VLR7 is selected as the new MA for that the user. The reason of updating









the HLR instead of the MA is that the cost of setting up and traversing the pointer chain

between MA and current serving VLR may be costly when the user is far away from the MA

and the connection setup delay for an incoming call may be intolerable. If an incoming call

arrives before the mobile user changes his or her MA, the current serving VLR becomes the

user's current MA because the HLR has the knowledge of the user's current location after

the connection setup and it is not necessary to go through the pointer chain to locate the

user again for the future service deliveries.

The call delivery procedure in the POFLA scheme is straightforward. When the sub-

sequent calls are initiated from some other switches to the user, the user's HLR is queried

first as in the basic procedure and a pointer to the user's potentially outdated MA is ob-

tained. The pointer chain is followed to find the user's current location. For example, in

Fig. 2.8, if the user is in RA6 when a call arrives. The user can be reached by following the

H pointer first and L pointer chain second. After the connection setup, the current serving

VLR becomes the user's new MA.

As we can see, in the POFLA scheme, the chain length can be longer than that in the

basic pointer forwarding scheme without increasing the connection setup delay significantly.

In section 2.5.3, we can see that the delay for the POFLA scheme is much less than that for

the per-user forwarding scheme. Our study also shows, under some assumption, the POFLA

scheme performs better than the static and dynamic local anchor schemes proposed by Ho

and Akyildiz [22]. Although in section 2.5.3, the two-level pointer forwarding scheme [45]

can generate similar results as that of the POFLA scheme, the new scheme is simpler to

implement than the two-level pointer forwarding scheme in practical systems.
2.5.2 System Model and Cost Functions

In this section, we develop an analytic model to derive the cost functions and compare

the performance of the POFLA scheme with IS-41/(;S\ MAP, pointer forwarding, local

anchor and two-level pointer forwarding schemes.

The mobile users in a PCS can be characterized by their call-to-,,,nl.:l.:l.u ratios (CMRs).

All the above schemes are evaluated under difference CMRs in this chapter. We observe

that a mobile needs to update its location only when the mobile does not engage any com-

munication with other users. Hence we can only need to compare the signaling traffic in the






26

time interval between call services (i.e., the interval between the end of the current call and

the beginning of the next call, which is called inter-service time in [49, 50]). Assume that a

mobile crosses a number of RAs during inter-service time. By ignoring the busy-line effect,

the inter-service time can be approximated by the inter-arrival time of calls to the mobile. If

the basic user location update scheme (IS-41) is used, the user's HLR will be updated every

time the user moves to a new RA. In the POFLA scheme, the HLR is updated only every

K1 K2 moves (K1 and K2 are the L pointer chain threshold and H pointer number limit,

respectively), and pointers are set up for all other moves. We define C' to be the total costs

of updating the location information (location update) and tracking the user (call delivery)

during the inter-service time for the the POFLA strategies. For convenience, we list all

notations used in our analysis as follows:

K1: the threshold for the L pointer chain;

K2: the limit of H pointer number (i.e., every K2 updates to an MA will result in the change
of a new MA);

m: the average cost of location update to the HLR;

M: the total location update cost during the inter-service time in the IS-41 scheme;

F: the total cost of call delivery in the IS-41 scheme;

M': the total location update cost in the POFLA scheme during the inter-service time;

F': the total call delivery cost in the POFLA scheme;

S1: the pointer setup cost of an L pointer;

S2,j: the pointer setup cost of the jth H pointer;

Ti: the cost of traversing an L pointer between two .,li.11:ent VLRs;

T2,j: the cost of traversing the jth H pointer;

a(i): the probability that there are i RA crossings during the inter-service time;

P: the processing cost of setting up a pointer (H or L);

G: the signaling cost of setting up an L pointer;

3: the coefficient of signaling cost for an H pointer (3 > 1);

p: the user Call-to-mobility ratio.








Then, we can express the total costs during the inter-service time for the POFLA scheme as

follows:


C' = M'+ F'. (2.17)


Since pointers are set up in the POFLA scheme, we need define the pointer setup and travers-

ing costs for further analysis. Every time a pointer is set up, the signaling messages will

be transmitted back and forth; For pointer travi -i,. the signaling message is transmitted

only in one direction. So we define the costs of pointer setup and traversing for the L pointer

as S1 and TI, respectively:


Sl = G+ P, (2.18)
1
Ti G + P. (2.19)
2

The costs of H pointers are not fixed because the length of the H pointer changes with the

user's mobility. In this chapter, we express the costs for H pointers as follows:


S2,j 0 if j 0 (2.20)
jG + P Otherwise

T 0 if j = 0
={ (2.21)
ljG + P Otherwise

where the j means the setup cost or the traversing cost for the jth H pointer.

Now, we can derive the formulas for M' and F' as follows: suppose that a user crosses i

RA boundaries during the inter-service time. The HLR is updated [L 2 ] times. If we call

the summation from the Oth to the (K2 )th H pointer setup cost, ZK -J S2j, the MA

update cost, then, there are [K ] times of such MA update costs that would incur during

the inter-service time with i RA crossings. In addition, there are [i-jLK1 K2 H pointer

setups and i [ti L pointer setups occurred in the remaining i RA crossings. Thus, we

can obtain
i-[ K1K2
21 L K,1 2 J
ii i
M' { +Sm + (i- L-)SI + [- 1( S2,) + S2, i )
i= j=0 j=0
(2.22)









The cost F' can be derived straightforward. In order to reach the user's current location,

the signaling message is sent to the MA and then travels through one H pointer (if any) and

i [ JK1 L pointers before reaching the current location. So, we have

K1
FP' = F+ {T2,( + (- [-]jKi)TI}a(i), (2.23)
i=0
li-LK2JK !K
where (= [L Kj j 2 Notice that we can easily obtain

K2-1 K2(K2 1)G + 2(K2 -)P
SS2,j -2
j=0

and
i- K12 ]K1K2

L S2,j Ki K1 ] 2 K1
j=0

So (2.22) can be rewritten as follow,

Mo / (K2 1)K^2G + 2(K2 I)P 2m
slM' a i(i) ] S L-a(,i) + 1)L ]a(,2 )
i=0 iO i=0
M1 M2 M3
G I L K2 ]K K2 2 +G + 2P [ i a- (2.24)
2 2 i-2Kc(i) 2 K1 (i) L(2.24)
i=O i=0
M4 M5

My can be simplified from the definition of a(i),


M, = S,1 G+P (2.25)
P P

According to (2.6) and applying variable substitution i = jK + k, then we obtain


a (jKII + k) (1- g)2 (K1)jgk yz j_ ,
pg

where
(1 g)2 K
y z g x= g.
pg









Thus, we have


oo K1-1
yS1E E jz xk
j=0 k=0

ySA(1- zKIx K(i
1-x
j=0
( g)(G + P)gKl-1
p(l gKi)


Similarly, we can obtain 1i, by using substitution i


a(jKIK2 + k)


(1 (gKIK2)gk
pg


where


( g)2
pg


(K2 1)K2PG + 2(K2


2

(K2 1)K2G + 2(K2 )P 2m
2


In a similar manner, we can also obtain M4 and 1 -, as follows:


Go K1K2-

PG- C


J2 k
j=0 k=0
j oo k 0


1 2


k 2
L-K-J UK Zk k)


K1


So-o K2-1 Ki-1
E (E y J( l2x nK+m)
j=O n-0 m=0
G(1 g) + g2
2pg(l gK1K2)( gKi)2
+(2K2 2K2 )gK(K2+) (2 1)


(2.26)


jKiK2 + k, then


yzixk.


z K1K2


x g.


1)P + 2m


I aLKwK (()
i=0

(1 g)gK1K2-1
p(l gKiK2)


(2.27)


K g KK2


(2.28)


'I -I .









and


G 92P K1K2-1
Y2 S L-J'(jKiK2 +k)
j=0 k0
PG + 2P KK2 k
2 KS L ZJXk
j=0 k-0
rG 2P K2-1 Ki1 -
G 2P y zJ 5 YnK'
j=0 n=0 m=0
(PG + 2P)(1 g)[(K2 l)gKi(K2+) K2gKK2 K+
2pg(1 gK1K2)(1 gK) )


Finally, we find the expression M' M f_ + M. + M4 + IfT2,( 0, we can

F' in the following fashion


F' =F + 2 i- P G + P + (i LK )(-G+ P)}(i)
i2L K1 K1J 2
i 0

SF+( G+P) ia(i)-( G+P)KI [L-J-Ia(i)
2 Y 2 K1
i=0 i=0
F1 F2
1 _0 i- [ ]KIK2
+ ( LG K1K2 + P)(i) .
+( 2 YC L K,
i=0
F3


obtain


(2.30)


Notice that if we use the substitution i = jKIK2 + k, when k = 0, 1, K1 1, T2, = 0.

So we can obtain F' as follows:

+ 2P K1(1 g)gK-l1 P( g)(gK-l KK2-1)
F' F+ [1 ]+
2p 1 gKi p( gK1K2)
/3G(1 g)[(K2 )g,(K2+1) K 1gK2 + gi]
(2.31)
2pg(1 gK1K2)(1- gKi)

In summary, we obtain

Theorem: If the inter-service time is I ,'!':. 1/,l,;//.7 distributed and the RA residence time

is generally distributed with Laplace transform f,*(s), then the total location update cost and








total cost for call delivery are given by

C + P (1 g)(G + P)gK -1
P p(l gKi)
(K2 1)K4G + 2(K2 I)P 2m (I g)gKK2-1 3G(l g)
2 p( gKK2) 2pg (1 KK2)(I_ gK)2
{gK1 + g2 K2gK1K2 + (2K2 2K2 )gK(K2+1) (2 1)2gK(K2+2)
(PG + 2P)(1 g) [(K2 1)gK(K2+1) K2K1K2 + gK
2pg(l gKK2)( gK)
SF G +2P KI(l g)gK-1 P(I g)(gK-l1 gKK2-1)
F = F [1- ]+
2p 1 gK p( gK1K2)
3G(I g)[(K2 I)gKi(K2) K2gKK2 +gK
2pg(l gK1K2)(1- gK)

where g = f*(A).
2.5.3 Performance Evaluations

We consider the situation when the RA residence time is exponentially distributed.

In our analysis, we do not address issues regarding the contents of messages and other

information transfer which may occur during a call connection setup. For simplicity, we

assume that the message sizes are equal for all signaling transactions. Since we only compare

the relative performance of the aforementioned schemes with the POFLA scheme under

various CMRs, the conclusions will not be affected by this simplification.

Notice that, in the simplified IS-41 or CS\! MAP procedures, the location update and

call delivery involve the same number of messages between HLR and VLR databases, so we

choose m = F. Without loss of generality, we normalize m = F = 1. G is the signaling

transmission cost and P is the processing cost. P usually includes the database transaction

costs. The values of P and G should be much less than m or F. In our works, we do not

assign any practical meanings to these parameters, they can be explained as the signaling

message traffic exchanged during the location management procedures or the time delay

experienced in the real systems.

In Fig. 2.9, we plots the relative location update, call delivery and net costs of three

schemes as functions of CMR. Here for the POFLA scheme and the two-level pointer

forwarding scheme, we assume K1 = K2 = 3; for the per-user forwarding scheme, the

threshold is nine (K1 x K2). In this figure, we also assume P = 0.05, G = 0.1 and 4 = 1.5.

As we can see in Fig. 2.9(a), the POFLA scheme generates higher values than the per-user







32

(a) Location update cost (b) Call delivery cost (c) Net cost
028 1 4 1
POFLA \ POFLA POFLA
Two level Twolevel Twolevel
-Per-user 1 Per-user Per-user
026 135 09

13- 08
0 24

125- 07
022-
12 \ 06
02-
1 15- \ 05

0 18 -
S11- 04

0 16105 03\
\'\ 105- \ 03-


014- 1 02
10 2 10 102 102 100 102 102 100 102
Call-to-mobility ratio Call-to-mobility ratio Call-to-mobility ratio

Figure 2.9: The relative costs for the three schemes with P = 0.05, G = 0.1 and = 1.5


forwarding scheme. It is obvious because in the latter scheme, only the L pointers are set

up while in the POFLA scheme, a new VLR would set up an H pointer to the MA when

the threshold for L pointers is reached, which costs more than an L pointer. For the two-

level pointer forwarding scheme, the level-2 pointer is the L pointer and the level pointer

is usually shorter than the H pointer [45]. So the location update cost for the two-level

pointer forwarding scheme is in the middle of them. Although, with the same length of the

pointer threshold, the per-user forwarding scheme can generate less location update cost,

it has the largest call delivery cost among the three strategies (see Fig. 2.9(b)). For some

users with small CMR, which means that the users have higher mobility relative to call

arrival, the call delivery cost for the per-user forwarding scheme is much higher than those

for the other two schemes. In practical systems, it could be embodied as the delay the users

have to wait before any connections can be set up. In the POFLA scheme, normally fewer

pointers have to be traversed than the two-level pointer forwarding scheme before a user

can be located, so the POFLA scheme has the least call delivery cost. In Fig. 2.9(b), the

performance of the POFLA scheme dose not improve much comparing with the two-level

pointer scheme; however the POFLA is easier to implement in practical systems. Although

the three schemes perform differently in location update and call delivery, the total net cost

for the three schemes are similar for high CMRs (Fig. 2.9(c)). In Fig. 2.10, we increase








33

(a) Location update cost (b) Call delivery cost (c) Net cost
032 14 1
-POFLA \ POFLA POFLA
Two level Two level Two level
03 Per-user Per-user Per-user
135 09
0 28
13 08

026
125- 07
0 24
12- \ 06
022-
115- < 05 -
02

S11- 1 0 04
0 18 -

016- 105- 03

014 1 02
10 10 10 10 10 10 10 10 10
Call-to-mobility ratio Call-to-mobility ratio Call-to-mobility ratio


Figure 2.10: The relative costs for the three schemes with P = 0.05, G =0.1 and = 3


the signaling transfer coefficient 3 from 1.5 to 3, which means that the H pointer setup

cost is higher. Under this conditions, both the location update and call delivery costs

for the POFLA scheme and the two-level pointer forwarding schemes increase. Since the

pointer setup and traversing costs for L pointer keep unchanged, the performance of the per-

user forwarding scheme does not change either. Even the costs of H pointer increase, the

connection setup costs of the POFLA scheme and the two-level pointer forwarding scheme

are still less than that for the per-user forwarding scheme (Fig. 2.10(b)). In Fig. 2.10(c),

when the CMR is very low, the per-user forwarding scheme can perform better; when the

CMR is larger than one, the net costs for the three schemes are similar. Fig. 2.11 shows the

performance for the three strategies when the pointer processing cost P is increased to 0.1.

With the increase of user population and the users' mobility, the processing cost for pointer

management would increase too. The processing cost includes database transactions and

may generate extra delay with larger number of operational requests. As we can observe

from Fig. 2.11, the costs for all three schemes increase. The POFLA scheme will generate

the least connection delay and the total net performance is very close.

In this chapter, we also compare the performance of our POFLA scheme with that of

the local anchor scheme. The authors [22] have -ii..-. -1. .1 two variants of the local anchor

scheme-the static one and the dynamic one. The static local anchor scheme is easier to



















(a) Location update cost
032
POFLA
Two level
S Per-user
03


10 10 10t
Call-to-mobility ratio


Figure 2.11: The relative costs for


(a) location update cost

POFLA
DLA


(b) Call delivery cost
17
POFLA
Two level
Per-user
16


(c) Net cost


05

12
04


1 1 0 3


1 02
102 100 102 102 100 102
Call-to-mobility ratio Call-to-mobility ratio


the three schemes with P = 0.1, G = 0.1 and = 1.5


(b) call delivery cost

POFLA
\ DLA


(c) net cost

POFLA
- DLA


03-
1 05

0 25-


02


015 1
102 100 102 102 100 102
Call-to-mobility ratio Call-to-mobility ratio


Figure 2.12: The relative costs for the POFLA and DLA


06



05



04



03
102 100 102
Call-to-mobility ratio


with P = 0.05, G = 0.1 and = 1.5










The net cost
POFLA
SStatic local anchor
1 8 -

16 -

14-

12



08-

06-

04

02-


10-, 100 10' 102
Call-to-mobility ratio

Figure 2.13: The relative net costs for the POFLA and SLA with P =0.05, G = 0.1 and
3 = 1.5


implement. In this scheme, a VLR serving a user is selected as the local anchor for that

user and will not change until the next call arrives. In the dynamic local anchor scheme, the

user's current local anchor might change to the current serving one according to the user's

expected future events. The dynamic local anchor scheme is more difficult to implement than

the static one; however the results in [22] shows that the dynamic scheme can guarantee that

the net cost is less than that of the basic IS-41 or CS(\ MAP strategy, and the static scheme

might generate higher cost than the basic scheme under some conditions. In fact, the local

anchor scheme is a special case of the POFLA scheme. If we let K1 = 1, then the POFLA

scheme reduces to the dynamic local anchor scheme. The performance comparisons of the

dynamic local anchor with the POFLA scheme are shown in Fig. 2.12. In order to make

the comparison fair, the effective pointer chain length (K1 x K2) in the POFLA scheme is

same as the dynamic local anchor scheme. In [22], the decision of the local anchor change

is made based on the user's next event, which is derived according to the user's mobility

pattern. In our analysis, we assume the local anchor changes to make sure the net cost will

not exceed the basic IS-41 scheme cost. In Fig. 2.12, we assume P = 0.05, G = 0.1, 3 = 1.5

and m = F = 1. Based on these assumption, we obtain the effective pointer chain length is

four, so we set K1 = K2 = 2. It can be seen that in both the location update, call delivery








36

(a) location update cost (b) call delivery cost (c) net cost
028 125 1
=0 01 -- O0 01 y=0 01
y=l1 l r= 11
0 26 -y=100 00 09 y=100
026
12
08
0 24

07
1 15
022
06-
02-
05




0 16\ /



102 10' 102 102 10' 102 102 10' 102
Call-to-mobility ratio Call-to-mobility ratio Call-to-mobility ratio


Figure 2.14: The effect of variance of residence time (7) with P = 0.05, G = 0.1 and 3 = 1.5


and the total net cost, the POFLA scheme has better performance than the dynamic local

anchor scheme. In Fig. 2.13, we also compare the total net costs of the POFLA scheme with

that of the static local anchor scheme. In this figure, we assume K1 = K2 3. We can see


that when the CMYR is low, the static local anchor scheme involves very high traffic load.

2.5.4 Sensitivity to the RA Residence Time


We now investigate the performance sensitivity of POFLA scheme to the variance of
014-------- I ---- ^ ^ 02--------





















the RA residence time. We assume that the RA residence time has a Gamma distribution.


For a Gamma distribution, the variance is V = i.e. a large 7 implies a small variance.

Fig. 2.14 shows the effect of 7 on '/IM, F'/F and C'/C, respectively. In these figures,

we can observe that the increase of the variance of RA residence time (smaller 7) causes

the increase of a /M and the decrease of F'/F; the net effect to CFICB is not significant.


The large variance implies that the number of RA boundaries the user crosses during the

inter-service time would vary greatly. If the user crosses many RAs, a longer H pointer

chain will be created. When the limit of H pointer K2 is reached, the HLR will be updated,

resulting in increase of Ma. On the other hand, if fewer boundaries are crossed, only shorter

pointer chains are set up, the pointer creation/tracing cost will be saved. The net effect is an


increase in M'. In Fig. 2.14(b), when the variance is high, if the crossed boundary number









is small, the call delivery cost is reduced; If the number is large, the pointer chain could be

shortened by H pointer or the update to HLR. The net effect is a significant improvement

in F'/F. The net effect of the residence time variance on the total cost ratio C'/C is not

significant for all CMRs (see Fig. 2.14(c)).
2.6 Discussions

In the two-level pointer forwarding and POFLA schemes, a pointer chain can be short-

ened by the MAs the user visited. Thus, the updates to the HLR is mitigated at the expense

of the increase of the local signaling traffic. It is beneficial when the cost of communicating to

HLR is relatively higher than the local signaling cost. One advantage of the two new pointer

forwarding strategies is that they can keep the FIND p.' n.i1ll low while significantly reduc-

ing the total system cost at the same time. As we can see from the previous sections, the

per-user forwarding and local anchor schemes are the special cases of the two-level pointer

forwarding and POFLA schemes. If we adjust K1 and K2 properly, the two new strategies

reduce to the per-user forwarding and local anchor schemes. There are some other ways to

set up the pointer chain. For example, all the wireless communication service areas can be

divided into Mobility Regions (\!lls). A user will update his/her location to his/her MA in

that region. Only when the user moves out of the region is a pointer set up from the old

MA to the new one. For the 3G wireless communication systems, 3GPP 23.119 specification

proposed an approach to limit the signaling traffic between the visited mobile system and

the home mobile system [9,51]. A new entity gateway location register (GLR) is introduced

between the VLR/SGSN and the HLR. From the viewpoint of the VLR/SGSN at the visited

network, the GLR is treated as the roaming user's HLR located at the home network. From

the viewpoint of the HLR at the home network, the GLR acts as the VLR/SGSN at the

visited network. Indeed, in the 3G wireless systems, a new level of location management

database is added. The users need to exchange extra local messages but reduce the long

distance or international messages exchanged in each of the subsequent registrations. The

two-level pointer forwarding and POFLA strategy can be implemented in the 3G systems in

the following manner: the GLRs can be selected as the MAs. A realistic implementation of

the new pointer forwarding schemes should also take into account the possibility that loops

may form as the user visits several RAs in succession. Thus, if a user revisits an RA and






38

a pointer for that user is found in that VLR, then the old pointer can be deleted to avoid

unnecessary operations. This is called "implicit pointer comprn --i"n in [23].
2.7 Conclusions

In this chapter, we propose two new pointer forwarding location management schemes

two-level pointer forwarding and POFLA strategies, which intend to reduce the cost of

location management by localizing or distributing the signaling traffic and to overcome the

HLR bottleneck problem while reducing the call setup (finding) delay. The performance

analysis is carried out to show the advantages of the new proposed schemes. Comparison

studies with the per-user pointer forwarding scheme and the local anchor scheme are also

undertaken and show that the proposed schemes outperform either one of them. More

importantly, the proposed schemes incorporates more parameters to be used to optimize the

performance of the location management. Moreover, the proposed schemes can be easily

tailored for the 3G wireless systems in which gateway location register is introduced.














CHAPTER 3
USER PROFILE BASED STRATEGY


3.1 Introduction

In the third generation wireless communication system, the high speed, real time and

multimedia services will be provided. The current RAs in the second generation systems

will be partitioned into smaller areas in order to meet the QoS requirements. It is obvious

that the small RA size will exacerbate the network location management signaling burden

significantly too.

It can be observed easily that most mobile users follow some fixed routes everyday. For

example, a person drives to his office every morning along a road he usually chooses and

-I .i, in office for most of the day and then comes home after work; A mailman delivers mails

along fixed itinerary everyday. If the network stores the mobile users' daily route information

in the profiles, then the location update signaling traffic can be minimized. Based on this

idea, a Profile Based Scheme (PBS) was proposed in [52, 53]. In this scheme, a user's daily

routine information is stored in the profile. If the user follows his or her itinerary well then

no location update message is sent so that the update traffic is reduced. When a call arrives

for that user, all the RAs the user could be in will be paged. The paging can be implemented

in all the RAs at the same time or operated one by one following a descending probability

order. The latter option is ;-.---- -1. .1 in [53]. It is straightforward that the location update

cost is reduced at the expense of increasing the total paging costs. A user who follows some

daily routine may deviate from his or her usual course because of road traffic, weather or

other reasons. If this happens, the mobile terminal is required to report to the new VLR

every time just like in the IS-41 or CS\! MAP scheme. So the mobile terminal needs to

store the RA list information in memory too and to be updated or adjusted by the network

periodically.

In this chapter, we propose a new scheme which is called user Mobility Pattern Based

Location Scheme (\! PBS) to improve the system location management performance. In the
39









new scheme, the User _1 [I '.:1.:1;7 Pattern (UMP) is considered and retrieved for location update

and call delivery procedures. The simulation results show that the MPBS generates less

signaling traffic than the IS-41 and PBS schemes and the paging delay in the MPBS is also

less than that in PBS. In the next generation wireless communication systems, the service

providers want to provide users the user-oriented services, namely, the system resources

are allocated according to the user's behaviors [54-56]. This will require the UMP to be

considered carefully. For example, in the MPBS scheme, the system can predict a user's

future location so that resource can be assigned in advance if the user is engaging in some

important applications. These procedures can improve the QoS greatly. In this chapter, we

give the detail description about the MPBS scheme and compare the performance with IS-41

and PBS scheme by simulations. In [53], the author did not consider the Call-to-Mobility

Ratio (CMR) effect on PBS performance. In our works, we also show that the user CMR

is very critical to the PBS and traffic saving can only be achieved under some limited CMR

range.

This chapter is organized as follows. In section 3.2, we describe the system architecture

to implement our scheme and introduce the VMN concept; Section 3.3 gives the details of

the PBS and MPBS schemes; We analyze the costs and present the simulation configurations

for the new scheme in section 3.4; Section 3.5 presents the simulation results and compares

the scheme performances under various conditions; The conclusion is drawn in section 3.6.
3.2 System Description

For the next generation wireless multimedia networks, different kinds of users have

different service requirements. Different services and QoS can be assigned to users based on

their past call or mobility history and their willingness of paying for a higher QoS. In order to

attract users, the network needs to store some user information in his/her profile and retrieve

it to find out what kind of service can meet the user's requirement. If the system stores

the necessary information in the profile, some system resource can be reserved in advance

in order to provide the user real time service. Furthermore, if the user mobility pattern is

known, some specific information can be prepared for a specific user in some specific area

the user will be in soon. If the above works, like information collection and dissemination,

are incorporated into the current distributed wireless communication database system, the









signaling and database retrieval traffic will overwhelm the system process capacity. The

delay resulted from the heavy traffic load will degrade the whole system performance. In

order to solve the problem, we propose the Virtual Management Network (VMN) concept.

VMN is an overlay management network on top of the existing wireless mobile network,

which handles the intelligent features. In our scheme, a new function entity, Mobility Agent

(\! A), is proposed. If we assume the signaling network for a wireless mobile network is

a Common Channel Signaling (CCS) network, like Signaling System No.7 (SS7), the MAs

can be hardware-based or software-based and connect to both VLRs and HLRs. At the

same time, the MAs are connected to each other. Each MA is in charge of a selected set of

location areas and can be dynamically configured according to the up-to-date traffic burden.

In SS7, the MAs can be placed together or separately at the same level as LSTP or RSTP.

If both LSTP and RSTP are equipped with MAs, we can say that the VMN is configured

hierarchically. The VMN can also connect to OA&M center.
3.3 PBS and MPBS Schemes

In this section, we describe the MPBS procedures in details; we also introduce the PBS

procedures generally and the details about the scheme can be found in [53].

3.3.1 Profile Based Location Scheme

The profile based location scheme (PBS) was proposed by WINLAB, Rutgers University.

In this scheme, the system maintains records of each user's most likely itinerary list. It is

assumed that the user location distribution probability is known in advance. It can be

provided by the mobile terminal or estimated by the system according to the user's past

calling history. The location list is stored in the switch that will conduct the search for the

user and the information required to update the list is stored within the mobile switching

center's billing records.

The user's itinerary can be defined as follows. If Ai is one of the location areas in the

record list, the user's most likely itinerary can be defined as {Ai} where k is the element

number in the set. The probability of a user being in a location area Ai is ai. The system

maintains a list of (Ai, ai) pairs for some time interval T. The probability of a user being

out of {A} 1, is given as
k
K -I ai. _(3.1)
i=1









If K > 0, there is always a probability of not finding the user. If the user follows his/her

daily itinerary strictly, namely, the user keeps roaming in {Ai} 1, no registration is needed.

When the user is out of the list, the mobile terminal is required to manually register to the

system. So the terminal must keep a copy of the list. The list can be sent to the terminal

by system and updated only when some Ai is added or deleted. When a call arrives for the

user, the location areas in the list are paged in descending order of a, until the user is found.

Under this strategy, the database will know a user's exact location when it is out of {A i} .

Indeed, when the user is out of the list, the PBS scheme performs the same as IS-41 or C(;S

MAP. According to procedures, The PBS can reduce the location update cost effectively

at the expense of increasing paging cost or paging delay. In [53], the authors only studied

the paging delay and the radio link cost under different list length. The paging delay was

derived, given three known probability distributions, in the term of the expected location

area numbers needed to be paged before the user can be found. In fact, it is intuitive that

the total cost of PBS scheme has tight relationship with the user's CMR. With small CMR,

which means the user has relative higher moving rate than call arrival rate, the PBS can

reduce most update cost and the final result is good. While, for the users with high CMR,

the paging cost will be dominant and the total cost of PBS may exceed that of the basic

IS-41 or CS\! MAP scheme.
3.3.2 Mobility Pattern Based Scheme

In order to improve the PBS scheme performance, we propose a new mobility manage-

ment strategy-mobility pattern based scheme (I\!PBS) in this chapter. The MPBS strategy

can reduce the user update cost and try to limit the paging cost at the same time. Compar-

ing with PBS, only two more elements, the time a user entering Ai and the residence time

in Ai, are added in the user profile. The MA will keep a list of 4-tuple (Ai, ti, Ti, ai) for each

user. We assume the cardinality is k. In the list, the tuples is not ordered according to a,

but to ti. For example, if a user's profile includes all the location entries he/she may visit

in 24 hours, the list is sorted by the time the user visits every location area. So for i $ j,

Ai and Aj may be same. In MPBS scheme, we define ; the user out-of-pattern probability,

which is given as
k
I. (3.2)
i=1








t t2 t tt5 t6
Ti1 2 T3 T4 T5













SOf / fice Restaurant/


Figure 3.1: Mobility pattern based scheme procedures

If ; > 0, there is probability that the user moves out of the location areas in the list. We

say the user is out-of-patter if this happens; otherwise, we say the user is in-the-pattern.

Fig. 3.1 shows an example of the MPBS. If a user lives in location area A1 and works in his

office building in A4. Before the user reaches his office, he will pass through A2 and A3; The

user also likes to have his lunch break in a restaurant located in As. If the user follows the

itinerary everyday and the information is stored in his profile, the network can locate the

user based on the profile and current system time.

To make the scheme clear, we need to define the user behaviors more precisely. In

the MPBS scheme, when users follow the UMP, the location update traffic can be reduced.

There are two kinds of patterns for users to follow in the MPBS scheme. When user enters Ai

at time t, and the residence time in Ai is T,, we say the user follows time-sequence pattern.

If the user enters and exits location areas following the A, order in the profile only, we say

the user follows the sequence pattern. It is obvious that a user following the time-sequence

pattern must follow the sequence pattern as well, but a user following the sequence pattern

may not follow the time-sequence pattern. However, users may deviate from their daily

routines because of weather or road traffic situations. So we need to find out how close a

user follows his/her mobility pattern. The mobile terminals for the next generation systems

should incorporate more intelligent functions. When a user roams in the network service

areas, we assume that the mobile terminal can record the location area ID, location area









entrance and exit time. We define the user's actual path information as the User Actual Path

(UAP). The UAP can be used to update the UMP periodically. As the time elapses or user

crosses location area boundary, the mobile terminal can tell whether the user is following any

pattern or not by comparing the UAP with UMP. The UAP is in the same format as UMP.

If we do not consider the time information, the UAP can be expressed as {Bi} 1, where

m is the length of UAP. Without the consideration of time, we can compare the similarity

between UAP and UMP by edited distance [57]. We assume the regular movement of a

mobile user can be modelled as an edited UAP by allowing the following legal operations:

Inserting a location area L at position i of the UMP gives UAP: A1, A2, ,
Ai_1, L, Ai, Ai+l, Ak,

Deleting the location area Ai of the UPM gives UAP: A1, A2, Ai-1,
Ai+l, Ak,

Changing a location area Ai of the UMP to another location area L gives UAP:
A1, A2, Ai_1, L, Ai+l, Ak.

As a result, the edited distance between a UMP and a UAP becomes the sum of the weights

of the editing operations. If the edited distance is less than a threshold, we say the user

follows the sequence pattern, indicating the general moving intention of the user. For large

systems with complex network topologies, the calculation of the spatial weights can be quite

involved. How to assign and calculate the spatial weight exactly is out of the scope of our

discussion. For simplicity and without loss of generality, we can define the weight as follows:

The cost of inserting

1 L is the .,li.l.:ent location area of Ai
W, =
oo otherwise

The cost of deleting


D 0 A1, *, Ai-1 have already been deleted
WD
1 otherwise

The cost of changing

1 L is the .,li..:ent location area of Ai
Sotherwise
oo otherwise









Based on the above notations, we can define user behaviors more precisely. When a user

enters a location area Ai, the user is said to follow the time-sequence pattern if and only if the

following requirements are met: (1) Ai E {A}; (2) tiactual-ti\ < AT and (3) t-ti-Ti < AT,

where ti,actual is the actual time the user enters Ai, t is the current system time and AT is

the time pattern threshold. The first condition constrains the user in the profile, the second

and third conditions limit the user to enter and exit location area Ai within some time

threshold. If we assume a UAP is {Bi}71, the edited distance between UAP and UMP is

d(A1, A, ... Am, B1, B2, Bm) and the edited distance threshold is AD, we can say the

user follows the sequence pattern if: (1) A, E {Af}; (2) tiactual -ti > AT or t-ti -T > AT

and (3) d(A1, A2, -. Am, B1, B2, Bm)/m < AD. In MPBS scheme, users can be in one

of the four states when enter a location area A:

State 1: If a user follows the time-sequence pattern, we define the user in state 1.

State 2: if a user follows the sequence pattern, we define the user in state 2.

State 3: if a user does not follow any of the above two patterns but A E {Ai}, we define
the user in state 3.

State 4: if A {Ai}, we define the user in state 4.

The reason we define four states is that the system can invoke different paging mechanisms

for users in different states. A user needs to register to the system when he/she switches

states. In the MPBS scheme, we assume the user state information is included in the update

message, so the system can know the user current state every time the mobile terminal sends

update message. When a user is in state 1, 2 or 3, no registration or state message needs to

be sent if the user keeps the state unchanged. If the user keeps in state 4, the terminal will

update its location to the system every time the user enters a new location area. Then, only

in state 4, the user needs to update the location every time he/she crosses an RA boundary.

In state 1, state 2 and state 3, the user needs to send update message only when the states

switch. Then if we can collect the user daily routine information well so that the user has

large probability in state 1,2 or 3, the update cost can be reduced. In the MPBS scheme,

when a call arrives, the system can adopt different paging strategies based on the user's

current state. If the user is in state 1, the system can decide which location area the user

is in according to the current system time and page it. If the current time is t and the user










60 1
-MPBS
IS-41
PBS
50 -


S40
0


E





10
20







0 1 2 3 4 5 6 7 8
Time (second) 104


Figure 3.2: The update numbers for three schemes in 24 hours with user residence time 30
minutes


is in state 1, the system can retrieve the tuple matching ti < t < ti + Ti, then the user can

be found in Ai. There is the probability that the user just moves out of the paged area and

enters the next one. If the users does not change state, only the next location area needs

to be paged. If the user is in state 2, the system knows the location areas the user is not

in according to the last location update. Since the user follows the sequence pattern, the

system knows the user must be in one of the location areas after the last updated one in

the profile. Because there is no time information in state 2, all the location areas the user

could be in will be paged in the descending order of the probabilities ai. If the user is in

state 3, the system knows that the user is in {Ai} and all the location areas in {Ai} will

be paged according to the descending order of a, until the user is found which is just like

the PBS scheme. If the user is state 4, the system knows the user's exact location area and

pages. In fact, when the user is in the state 4, the MPBS scheme is exactly the same as

IS-41/C(;S\ MAP scheme. As we can see from the three schemes, the IS-41 will generate the

most update messages, the PBS scheme will generate the least ones and the MPBS is in the

middle of them. Fig. 3.2 shows this clearly. Although the MPBS generates more update

messages than the PBS, it reduces the paging cost dramatically than the PBS and achieves

total cost reduction.









3.4 Cost Evaluation and Simulations

In this section, we try to estimate the total signaling costs for the three location man-

agement schemes. Since the user behaviors and many other random factors will affect the

results significantly, it is hard to get the close form of the cost functions [58-60]. So, in

this chapter, we just give the general estimation for each scheme signaling cost. The scheme

performance comparison is carried out by simulations.

We define C, the cost for location registration and C, the cost for paging one location

area. In our analysis, we do not consider the call delivery processing cost because it is

same for all the three schemes. For IS-41/C(;\ MAP scheme, the total cost between two

consecutive calls can be given below [45]


CGSM Cp, (3.3)
p

where p is the CMR. In the following analysis, we assume there are k location areas in the

user's profile. For the PBS scheme, we define E(k) the average location area number that

has to be paged before the user is found. So it is straightforward that the total cost for PBS

scheme is

CU
CPBS = l1- + (1 ,z)E(K)Cp + Cp
p
CU
W= +i- [(1 E(k)) + E(k)]C, (3.4)
p

where wi1 is the probability that the user moves in and out of the profile between two

consecutive call arrivals in the PBS scheme. The analysis of total cost for MPBS is more

complicated. If we define Cp, the paging cost for users in state i and 7i the probability

that the user is in state i when a call arrives, respectively, the total cost of MPBS can be

expressed:
S4
CMPBS C= 7 i, (3.5)
P i

where c02 is the probability of the user's movements that need to update to the HLR. It is

obvious that 3i i = 1.

We need to derive Cp, to specify the total cost of MPBS scheme. In the MPBS, when

user is in state 1, if a call arrives, the location area the user is currently in will be paged








according to the profile. However, there is probability that the user just moves into the next

location area when the current location area is paged. In order to make sure the user can be

found, the next location area needs to be paged if no response is received in predefined time

in current area. We assume that tiactul ti follows Gaussian distribution with zero mean

and variance a. If ti,actual ti > AT, we say the user is not in state 1 anymore. Then the

probability that the user just moves to the next location area when a page message arrives

is 0.5 Q( -). Since the paging expiration time is very short, we ignore the probability

that the user crosses more than one location areas when the paging message arrives. Based

on the above results, we can obtain


G G + [0.5 Q( )]C

S [1.5 Q(AT)]C. (3.6)


If a user is in state 2 when a call arrives, only the following location areas need to be paged

because it is known that the user is not in the first m (m < k) location areas. So the

distribution probability in the following location areas is conditional. If we assume there are

m location areas has been crossed before a call arrives, and let Xm = a + a2 + + O

and k' = k m, we can get the conditional probability distribution for the next k' location

areas as a = -XM O2 -m a = kX respectively. If the average paged number

is E(k'), the Cp2 can be written as


C2 E(k')C,. (3.7)

In the real situation, it is may be difficult to get the conditional probabilities. But in MPBS

scheme, there is no necessity for the system to compute them because the system only needs

to page the following location areas in the descending order of am+, ,m+2, ak. Divided

by a same positive value does not affect their order. The cost of CP, is just like PBS scheme

and the cost of Cp4 is the same as IS-41/CS! I MAP scheme. So we have


Cp, E(k)C,, (3.8)

and


CP4 p- CP


(3.9)























Figure 3.3: Simulation network architecture


So, we can rewrite (3.5) as


CMPBS 2 + [(1.5 -Q( ))
P o-
++2E(k') + 3E(k) + 74]Cp. (3.10)

In our simulations, we treat the state update message the same as the registration

message and normalize the update cost C, = 1. The paging cost is usually less than update

cost, so we assume the paging cost for one location area is 0 < C, < C,. We are interested

in how the CMR, the paging cost Cp, the user distribution probability a, and the user

out-of-pattern probability ; affect the performance of the PBS and MPBS schemes. In this

part, we assume there are three representative profile distribution [53]: uniform, linear and

exponential. Denote the conditional probability of a user being in the ith location area in the

list as 4j = ai/(1 K) or p, = aj/(1 ;). The definitions for the three kinds of distribution

are given as below.

Uniform Distribution: when P1 = 2 '= k = /k, the profile is said to be uniformly
distributed.

Linear Distribution: when 3, = (k+li for i c {1, 2, -- k}, the profile is said to be
linearly distributed.

Exponential Distribution: when /3 =- 1' b b for i c {i 2, ,k}, the profile is said
to be exponentially distributed, where b is a constant.

The simulation network architecture is shown in Fig. 3.3. The network consists of

16 RAs. Each RA has 4 neighboring RAs. The links between RAs are virtual links, which

imply the user can roam bidirectionally between the connected RAs. The links between RAs







50




0.9

0.8

D 0.7


0.4
ao 0.6




0.2
S0.4-

c 0.3 -

0.2-

0.1


0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
The user Call-to-Mobility ratio

Figure 3.4: The location update cost ratio of MPBS scheme to IS-41/(;S\ MAP


and HLR are signaling links [61]. When the user enters an RA, depending on the schemes

currently adopted, an update message may be generated and sent to HLR. The user resides

in an RA for some exponentially distributed time with predefined mean, then moves into

the next RA [62]. The calls for that user generated in HLR forms a Poisson process. For

simplicity and clarity, we embed the VMN functions into the HLR node. The HLR node will

record all the update and paging costs for different schemes at the end of the simulations. In

the simulations, we assume the update cost for every location area is same. The simulations

are event driven. In the initiation, a user is generated randomly in one location area and set

in state 1. The state 4 is different from the other states. In state 4, the user in fact does not

follow any pattern. So when the user is in state 1, 2 or 3, we say the user is in-the-pattern,

and the user is out-of-pattern when he/she is in state 4. We also assume 90 percent of the

time, the user is in-the-pattern. The conditional probability of the user in state 1, state 2

and state 3 are 0.8, 0.15 and 0.05, respectively. All the simulations collect the user tracking

information for 24 hours. In order to study the out-of-pattern probability affection on the

scheme performance, the p will change during the simulations. For simplicity but no loss of

generality, we assume ai and ti are in the same order in the profile and Ai / Ay for i / j.

















E 0.38-

o 0.36

m 0.34
o) / i

00.32

0.3
--- Paging cost=0.1
-e-- Paging cost=0.5
0.28

0.26
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
The user Call-to-Mobility ratio

Figure 3.5: The comparison of the locating time for MPBS and PBS


3.5 Numerical Results and Comparison

As we mentioned before, the MPBS can reduce the location update signaling traffic.

Fig. 3.4 shows the relative location update cost of the MPBS to IS-41 or (;CS\ MAP scheme.

In this simulation, the out-of-pattern probability p = 0.1 and the probabilities of user in

three different states are set as above. In Fig. 3.4, the update cost of MPBS scheme is

less than half of the update cost in the IS-41/C;S\! MAP scheme, and the CMR does not

affect the performance dramatically. The update cost is usually larger than the paging cost.

That is how the PBS and MPBS can achieve total saving by reducing the update cost at the

expense of increasing the total paging cost. However the MPBS can reduce the total cost

without increasing the paging cost too much. In both PBS and MPBS schemes, the system

usually pages more than one location areas trying to find out the user's exact location. In

other words, the MPBS and PBS will introduce some delay during call delivery procedures.

In Fig. 3.5, we can see that the paging delay generated by MPBS is much less than PBS.

The reasons are follows. In PBS scheme, all the location areas in the list are needed to be

paged. Usually, the location areas are paged sequentially according to the user distribution

probability. So the delay is the time elapsed from the paging messages is sent to the first

location area to the user responses. The delay can be different with different probability

distributions. In Fig. 3.5, we assume the user profile distribution is uniform. In the MPBS

















-- MPBS with finding cost=0.5
PBS with finding cost=0.5
1.5 -
0








0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
The user Call-to-Mobility ratio

Figure 3.6: The total costs of MPBS and PBS to IS-41/CS_\! MAP with uniform distribution


scheme, when the user is in-the-pattern, only when the user is in state 3, need all location

areas to be sequentially paged. When user is in state 1, only one location area is paged in

most of time; when in state 2, only part of the {Ai} need to be paged. The paging cost

could be considered as the paging delay for one location area. In our simulations, we assume

the paging cost for every location area is equal. As we can see, when the CMR is low, the

paging delay for MPBS scheme is 70% less than PBS scheme. The reason is that when the

CMR is low, the user has relative high movement probability, the PBS scheme will page

more location areas trying to find out the user; but the MPBS scheme is not affected by this

factor. When the CMR is large, the user will stay in a location area for a relative long time,

the PBS scheme can find out the user with less location area paging. However, the MPBS

total paging cost is still 60% less than PBS scheme.

Although the MPBS has less paging delay than PBS, we need to examine the total

costs for the two schemes and try to see whether they can achieve better performance than

conventional IS-41/CSI MAP scheme or not. In Fig. 3.6, 3.7 and 3.8, we plot both MPBS

and PBS to IS-41/C(;~S MAP total cost ratios with three different probability distributions

and different paging costs. We can see from these figures, the costs for PBS scheme increase

very quickly with the increase of paging cost. When the paging cost is 0.5, the PBS scheme

can have saving only when the CMR is very low. In that situation, the user will make


















2.5





2


o





I.
1.5



0
0
01





o 0.5


0 0.5 1 1.5 2 2.5 3
The user Call-to-Mobility ratio


3.5 4 4.5


Figure 3.7: The total costs of MPBS and PBS to IS-41/GSM MAP with linear distribution












1.6


-- MPBS with paging cost=0.1
PBS with paging cost=0.1
S- MPBS with paging cost=0.5
PBS with paging cost=0.5







,it----i


0.5 1 1.5 2 2.5 3
The user Call-to-Mobility ratio


3.5 4 4.5 5


Figure 3.8: The total costs of MPBS and PBS to IS-41/CS_\ MAP with exponential distri-

bution


-0- MPBS with paging cost=0.1
S : -E- PBS with paging cost=0.1
S- MPBS with paging cost=0.5
PBS with paging cost=0.5





:


1.4
o
..Q

r 1.2







2 0.8
(5
0

o 0.6


0
o

S0.4



0.2









a lot of updates to the system and the update cost is dominant. With the increase of

CMR, the paging cost for PBS scheme plays a more important role, then the total cost

increases fast. The MPBS total cost increases much slowly with the paging cost. This is

the advantage of MPBS. In the wireless multimedia networks, the data service will consume

a lot of bandwidth, which makes the paging cost higher. The MPBS total cost increases

much slower than PBS does with the increase of paging cost. Another advantage of MPBS

is that the cost curves are flatter than PBS in a wide range of CMR when the paging cost is

large. That means the MPBS scheme is applicable to different classes of users with different

mobility pattern. In Fig. 3.8, the total cost of PBS scheme is a little less than MPBS scheme

when the paging cost is small. The reason is that, for exponential distribution, the paged

location areas is less than other distributions. When the paging cost is small, the total cost

may be small. But even with exponential distribution, the PBS total cost increase much

faster than MPBS scheme does with the paging cost.

In the above situations, we assume 90 percent of the time, the user is in-the-pattern. It

is intuitive that the total costs of both MPBS and PBS schemes have important relationship

with the out-of-pattern probability like K or c. Fig. 3.9 and Fig. 3.10 show how the user

out-of-pattern probability affects the system performance for both MPBS and PBS schemes

with different probability distributions. In both figures, we assume CMR = 1. The paging

cost is 0.1 in Fig. 3.9 and 0.5 in Fig. 3.10. In Fig. 3.9, both the PBS and MPBS schemes

total costs increase as the user out-of-pattern probability increases. The cost of MPBS is less

than PBS. When the paging cost is large, as in Fig. 3.10, the total costs for PBS are larger

than 1 for any one of the three distributions. This also proves the conclusion we made before,

the PBS scheme can improve the system performance only in very limited conditions. In

Fig. 3.10, the total costs of MPBS are still less than one even the user has large probability

to get out of the pattern. In the two figures, all the cost ratios are equal to one when the

user out-of-pattern probability is one. The reason is obvious. When the user out-of-pattern

is one, the user will not enter any location area in the profile list and needs to register to

the system every time he/she moves. The system mobility management scheme is the same

as IS-41/C(S\ scheme indeed.




































I IlF -.11 ,,,,,,,,I,:.,,,,, ,,.I
-- MPBS with linear dist.
-x- MPBS with exponentional dist.
PBS with uniform dist.
-x- PBS with linear dist.
PBS with exponentioanl dist.


0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
The user out-of-pattern-probability


Figure 3.9: The effects

cost = 0.1, CMR = 1


of user out-of-pattern probability on MPBS and PBS with paging


0 1.8



1.6



0 0


S1.2






0.8



0.6
0 0.1 0.2


-0- MPBS with uniform dist.
- MPBS with linear dist.
- MPBS with exponentional dist.
PBS with uniform dist.
S PBS with linear dist.
PBS with exponentioanl dist.


0.3 0.4 0.5 0.6
The user out-of-pattern-probabilit


0.7 0.8 0.9 1
y


Figure 3.10: The effects of user out-of-pattern probability on

cost = 0.5, CMR = 1


MPBS and PBS with paging


01
0I



S0.8-


0.7-

0

0
0
'"

0
o


0.4

0.3


~J~ i",









3.6 Conclusions

For the next generation wireless multimedia communication systems, the radio spectrum

is the scarcest resource. In order to exploit the radio resource and provide users services

more efficiently, the location areas become smaller. The result is that the user location

update message will consume a lot of bandwidth. This situation becomes worse with the

increase of user number. Many research works have been carried out to optimize the system

performance in user mobility management area, like the PBS scheme. In PBS, no update

message needs to be sent as long as the user is in the profile list. Only when the user gets

out of the profile, need registration messages be sent to the system. When a call arrives

for the user, the location areas in the profile are paged according to the descending order

of distribution probability, until the user is found. In the PBS scheme, the user location

update message cost is saved at the expense of increasing the paging cost. In this chapter,

we studied the PBS scheme performance based on different probability distributions, paging

costs and CMRs. The affection of the user's out-of-pattern probability is also investigated.

The results show that PBS scheme only works well for very small CMR and the total cost

increases quickly with the paging cost. We also proposed a new location strategy-MPBS

in this chapter. In MPBS scheme, the user mobility time pattern is recorded in the user's

profile too. When user is in-the-pattern, there are three states the user could be in. The

user updates his/her location only when the states or pattern change. The simulation results

-II--., -1 that, although the MPBS scheme generates more update messages than PBS does,

the total cost of MPBS is usually significantly less than PBS and the MPBS scheme is not

very sensitive to the increase of paging cost either. This is very important because the paging

operation will become costly in the multimedia networks where wide bandwidth services will

be provided. One of the most important QoS factors is the connection setup time. In our

system architecture, the setup time is embodied by the paging delay. Our simulation results

show that the paging delay of MPBS scheme is over 50% less than PBS scheme. Our results

also show that MPBS scheme can work well for user with different CMR. In this chapter,

we also propose the system architecture to implement MPBS scheme and a new node item,

Mobility Agent (\1.A), is introduced. The MAs form a new support network for the wireless

communication system and connect to the system databases and operation-control center.






57

The MAs can collect and process the users' history information in their charge areas and

provide the user more specific services.














CHAPTER 4
MOBILITY MANAGEMENT IN MOBILE IP NETWORK


4.1 Introduction

The current fast increasing demand for wireless access to internet applications is fueled

by the remarkable success of wireless communication networks and the explosive growth of

the Internet. The future generation wireless networks target to provide users with high-speed

internet access and more sophisticated services besides voice communication services [63].

The user equipment, such as wireless laptops, cellular phones and palm pilots, make it

possible for mobile users to access the internet applications that predominantly based on

IP technology [64, 65]. The popularity of the Internet provides strong incentives to service

providers to support seamless user mobility. However, many telecommunications systems

such as first and second generation wireless cellular systems were designed mainly for voice

services, the integration with data networks becomes the major push for third generation and

future generation wireless systems. Mobile IP is the mobility-enabling protocol developed

by the Internet Engineering Task Force (IETF) to support global mobility in IP networks

[11, 16]. This standard has become the solution to solve the user mobility in almost all

wireless mobile systems.

The IP protocol has been designed for wired networks. There are two major functions

for the terminal IP address in the Internet. An IP address is used to identify a particular

end system in the whole network and is also used to find a route between the endpoints.

In IP networks, the packets delivered to a particular end system are routed based on the

destination IP address by the intermediate routers. Based on this observation, we can

conclude that a mobile terminal needs to have a stable IP address in order to be stably

identifiable to other network nodes and also needs a temporary IP address for the routing

purpose. The Mobile IP protocol extends IP by allowing a mobile node to effectively utilize

two IP addresses, one for identification and the other for routing.






59

Mobile IP enables mobile terminals to maintain all ongoing communications with the

Internet while moving from one subnet to another. In the Mobile IP protocol, mobile

terminals that can change their points of attachment in different subnets are called Mobile

Hosts (I\!Ih). An MH has a permanent address (home address) registered in its home

network and this IP address remains unchanged when the user moves from subnet to subnet.

This address is used for identification and routing purpose, which is stored in a Home Agent

(HA). A HA is a router in a mobile node's home network, which can intercept and tunnel

the packets for the mobile node and also maintains the current location information for the

mobile node. If an MH roams to a subnetwork other than the home network, this subnetwork

is a foreign network for that user. In the current Mobile IP protocol, the MH can obtain

a new IP address from a router in the visited network. The router in the visited networks

which assigns MH the IP address is the MH's foreign agent (FA) and the new address is

the MH's care-of address (CoA) used for packet routing purpose. The CoA for the MH will

change from subnet to subnet. In order to maintain continuous services while the user is

on the move, Mobile IP requires the MHs to update their locations to the HAs whenever

they move to different subnets so that the HAs can intercept the packets delivered to them

and tunnel the packets to the user's current points of attachment. Thus, the Mobile IP can

provide continuous internet access services for mobile users and does provide a simple and

scalable solution to user mobility.

However, Mobile IP is not a good solution for users with high mobility. Its mechanism

requires every MH to update its new CoA to the HA every time the MH moves from one

subnet to another, even though the MH dose not communicate with others while moving. As

shown in Fig. 4.1, the location update cost in Mobile IP can be excessive, especially for the

mobile users with relatively high mobility and long distance from their HAs. This problem

becomes worse with the increase of the mobile user number [66]. Moreover, if a user is far

away from his/her home agent or the HA processing capability is overwhelmed by the huge

volume of update messages, the signaling delay for the location update could be very long,

which will result in the loss of a large amount of in-flight packets and degrade the Quality

of Service (QoS) [67].










HA -- Location registration
Packet delivery
MH movement
HA Home Agent
FA Foreign Agent
MH Mobile Host




INTER ET







I F
MH MH -MH

Figure 4.1: The MIP location registration and packet routing


In this chapter, we propose a dynamic hierarchical mobility management scheme (DHMIP)

for Mobile IP networks. In our scheme, the location update messages to the HAs can be re-

duced by setting up a hierarchy of foreign agents for mobility management, where the level

number of the hierarchy is dynamically adjusted based on each mobile user's up-to-date

mobility and traffic load information. Analytical model is developed for the performance

evaluation. Analytical results show that our new scheme outperforms the Mobile IP and

IETF hierarchical Mobile IP schemes under various conditions. The important contribution

of our research is the new approach we develop in this work. Most performance evaluation of

mobility management schemes in mobile IP networks is carried out by simulations. Our work

presents a novel analytical approach to the performance evaluation of mobile IP networks.

This chapter is organized as follows. We introduce the related works on the mobility

management for the Mobile IP networks in section 4.2; The detail procedures of the DHMIP

scheme are presented in section 4.3; In section 4.4, we develop an analytical model to derive

the signaling cost functions for the new scheme; The DHMIP scheme performance is demon-

strated in section 4.5. Section 4.6 compares our new scheme performance with that of the

IETF Hierarchy Mobile IP scheme and some improvements which can enhance the DHMIP

performance are detailed in section 4.7. Section 4.8 gives the conclusions.









4.2 Related Works

User mobility in wireless networks that support IP mobility can be broadly classified

into macro-mobility and micro-mobility. The macro-mobility is for the case that an MH

roams across different administrative domains of geographical regions. The macro-mobility

occurs less frequently and usually involves longer timescales [12]. The Mobile IP can ensure

the mobile users reestablish communication connections after a move during macro-mobility.

The micro-mobility means the MH movement across multiple subnets within a single network

of domain. For micro-mobility, which occurs quite often, the Mobile IP paradigm needs to

be enhanced. Most of the related works attempt to improve the Mobile IP micro-mobility

handling capability [13]. In [66], the authors proposed a scalable mobility management

scheme which uses hierarchical FAs to handle the user mobility within one subnetwork for

wireless internet, and FA hierarchy in this scheme is pre-configured. In this architecture,

the Base Stations are assumed to be network routers. The higher levels of the hierarchy rely

on Mobile IP to handle the macro-mobility. The Handoff-Aware Wireless Access Internet

Infrastructure (HAWAII) is a separate routing protocol to handle micro-mobility [40]. The

scheme hinges on the assumption that most user mobility is local to an administrative domain

of the networks. An MH entering a new foreign network is assigned a new CoA and retains

its CoA unchanged while moving within the foreign domain. In this scheme, the HA and any

corresponding host are unaware of the host's mobility within that domain. The route to the

MH is established by specialized path setup schemes that update the forwarding tables with

host-based entries in selected routes in that domain. For macro-mobility, the HAWAII uses

the traditional Mobile IP. In this sense, this scheme can be considered as an enhanced Mobile

IP. The Cellular IP scheme is introduced in [41]. In this scheme, the location management

and handoff support are integrated with routing in Cellular IP access networks. The network

is connected to the Internet through a gateway router and the roaming between gateways is

managed by Mobile IP while the mobility within access networks is handled by Cellular IP.

All the packets originated from or terminated to the MHs are handled by the gateways, and

the host location information is refreshed by the regular data exchange transmitted from

mobile host. The Cellular IP also supports IP paging. The Paging mechanism can minimize

the signaling traffic supporting the mobility management for users in the idle state. When






62

packets need to be sent to an idle mobile host, the host is paged and becomes active. To

localize the signaling traffic of supporting IP services in cellular networks, Das et al [12]

proposed to use the mobility agent (1\.A) to localize the update traffic, leading to a new

architecture, called TeleMIP. In the TeleMIP, an MA is in charge of one region, handling

the CoA addresses for those MHs roaming in the region. In the HA, the MA is registered.

Whenever an MH changes an FA to another FA in the same region, it only updates the

MA. When an MH crosses the region boundary, the MH registers with the new MA, which

then sends an update message to the HA. It has been demonstrated that TeleMIP does

enhance the performance in the IP support over the cellular networks. However, this new

architecture may add the network management entity and complexity. In the Mobile IP

networks, to reduce the registration signaling traffic (for binding update), the Mobile IP

regional registration is proposed in [42], which is called IETF hierarchical scheme in this

chapter. The protocol employs the FA hierarchy to localize the registration traffic. In this

protocol, the HA registers the publicly routable address of the Gateway FA (GFA) and

the MHs location update messages establish tunnels in a regional network along the path

from MHs to GFA. Although multiple levels of hierarchy are mentioned in [42], typically

one level architecture, where all FAs are connected to the GFAs, is used for discussions. In

the IETF hierarchical scheme, the network architecture is centralized. It is not clear and

usually hard to determine the size of a regional network. The mobile users up-to-date traffic

load and mobility may vary, and the fixed structure is lack of flexibility. To overcome this

deficiency, Xie and Akyildiz proposed a distributed dynamic regional location management

scheme for Mobile IP [43]. In this scheme, the first FA an MH registers at in a new regional

network is selected as the GFA, and the regional network size is adjusted based on the user's

current traffic load and mobility information. It can be considered as the extension of the

IETF regional registration scheme to make it more flexible and adaptive. In this chapter,

we propose another dynamic hierarchical mobility management scheme for the Mobile IP

networks. In our scheme, when an MH changes its subnet and obtains a new care-of address

from the new FA, the new FA updates the new address to the MH's previous FA so that the

new FA forms a new location management hierarchy level for that user. The FA hierarchical

architecture is specific for every user, which makes the user avoid updating his/her home






63

network frequently. The packets delivered to the MH can be tunnelled via the multiple levels

of FAs to the user. In order to avoid long packet delivery delay, there is an optimal number

(or threshold) of hierarchy level for each user according to his/her call-to-mobility ratio

(CMR). The threshold can be dynamically adjusted based on the up-to-date mobility and

traffic load for each terminal. When the threshold is reached, the MH updates its location

to the home network and sets up a new hierarchy for its further movements. The optimal

threshold for each user can be derived by an iterative algorithm. Since our scheme reduces

the registration traffic to the HAs significantly, it can also reduce the in-flight packet loss

greatly [68].

One significant contribution in this work is the analytical approach we develop for

the performance evaluation of wireless mobile IP networks. Most of past research focuses

on the system architecture for mobility management in mobile IP networks. Performance

evaluations are carried out mostly by simulations. In our work, we develop a systematically

analytical approach to evaluate the proposed mobility management scheme in mobile IP

networks. It is our hope that our work can open a new avenue for performance evaluation

of mobile IP networks.
4.3 Dynamic Hierarchical Location Management Scheme

In our new dynamic hierarchical system architecture, there is no fixed hierarchical ar-

chitecture for users or any restriction on the shape and the geographic location of subnets.

In the Mobile IP protocol, an MH can determine if it enters a new subnet by detecting

the agent advertisement messages sent by the mobility agents (HAs or FAs). The MH then

obtains a new CoA from the new serving FA and sends the location update message to its

HA. Upon receiving the message, the HA can set up a binding between the MH permanent

address and current CoA so that the HA can intercept the packets to this MH and tunnel

them to the user's current access point. The MHs in the Mobile IP networks are required

to update their new care-of addresses whenever they change the locations (subnets) even

though the MHs do not communicate with others. As shown in Fig. 4.1, this procedure

could result in heavy signaling traffic to the networks.

In our dynamic hierarchical Mobile IP scheme (DHMIP), the location update signaling

traffic can be reduced by registering the new CoA to the pervious FA as shown in Fig. 4.2.










HA
S Location registration
Packet delivery
MH movement
HA Home Agent
FA Foreigen Agent
MH Mobile Host




INTERNET




FA FA --FA3 FA -FA FA6


MH -M MH MH

K i

Figure 4.2: The DHMIP location registration and packet delivery


By this procedure, a dynamic location hierarchy is constructed for a specific mobile user. The

packets for this user can be intercepted and retunneled along the FA hierarchy to the mobile

terminal. Thus, the location update traffic can be localized. In this scheme, we can adopt

the similar procedures in [70, 71] to notify the previous FA the user's new CoA. However,

the packet forwarding by multiple FAs will cause some service delivery delay, which may not

be appropriate when there is delay restraint for some internet applications such as video or

voice services. In order to avoid excessive packet transmission delay, we set a threshold to the

hierarchy level number in the DHMIP scheme. When the threshold is reached, the MH will

register to its home agent. In the DHMIP scheme, the threshold is adjusted dynamically

based on every user's current traffic load and mobility. The performance of the DHMIP

scheme is shown in Fig. 4.2. In this figure, an MH moves from subnet1 to ,-,i,1;, 1,. We

assume the threshold level of the hierarchy is three. When the user is in subnet2, subnet3,

subnet5 or -,,1n, 1,.. the MH updates the new care-of addresses to the previous FAs. Since

the previous FAs are usually close to the new ones, the location update cost is less than

that to HA. After the user enters subnet4, the hierarchy level threshold is reached and the

MH will set up a new hierarchy. In this case, the MH updates its new CoA to the HA

directly. When the user is in subnets or -' /, I,.. there are packets arrivals for the user. The

packets are then intercepted by the HA and tunnelled to the user. Since the HA does not








Table 4.1: The dynamic hierarchical Mobile IP protocol

% Location registration procedures
Kopt: the optimal hierarchy level threshold;
Initialize i = 0;
if (\II enters a new subnet)
i =i+;
if (i < Kpt)
New FA registers to the previous FA;
else
New FA registers to the HA;
i 0;
Compute the new Kopt;
% Packet delivery procedures
if (packets for the MH are intercepted by the HA)
Tunnel the packets to the first FA;
if (the first FA is not the MH current serving FA)
Retunnel the packets to the current FA;
The current FA decapsulates the packets and sends them to the MH;


have the user's up-to-date location information, the packets are sent to the FA that the user

updated last time. In our example, they are FA1 and FA4 in Fig. 4.2. Thus, the packets are

retunneled along the hierarchy to the user. The optimal hierarchy level threshold Kopt can

be computed based on the user's current traffic load and mobility pattern. The Kopt can be

adjusted in different epochs in the DHMIP scheme. For example, the optimal value can be

updated every time the MH enters a new subnet or when the previously calculated threshold

is reached or the MH calculates it periodically. There is a trade-off between the accuracy of

Kopt and the MH's computational power consumption. The more often the update of Kopt,

the more accurate the value and the more signaling traffic saving; however the more power

consumption. The DHMIP scheme can be described by the pseudo-code in Table 4.1. In

this chapter, we assume the optimal value is updated every time the last optimal threshold

is reached.
4.4 Analytical Model

In this section, we develop an analytical model to derive the location update and packet

delivery cost functions for the DHMIP scheme. Just like in [43], we do not consider the

periodic binding update costs that an MH sends to its HA or FAs to refresh their caches.

We define the following parameters for our analysis in the rest of this chapter.

K : the crossed subnet number between the last packet arrival and the final location update
just before it (see Fig. 4.2);









p: the MH call-to-mobility ratio (CMR);

K : the threshold of the FA hierarchy level;

U : the average MH location update cost to its HA;

F : the packet delivery cost for a packet to MH in a foreign network under the MIP scheme;

M' : the total location update cost for an MH incurred between two consecutive packet
arrivals under the DHMIP scheme;

F' : the average packet delivery cost for a packet to MH in a foreign networks under the
DHMIP scheme;

T : the packet delivery cost between FAs in the DHMIP scheme;

S: the hierarchy setup cost in the DHMIP scheme.

Let a(i) denote the probability that an MH crosses i subnets between two consecutive

packet arrivals. Under the current Mobile IP scheme, the total location update and packet

delivery cost can be shown as:


C(p) = iUa(i) +F
i=0
+ F. (4.1)
P

In this chapter, the CMR is defined as follow: if packets arrive at an MH at rate A and time

the user resides in a given subnet has a mean 1//p, then the CMR, denoted by p, is given as


p = A/p. (4.2)

The cost for the DHMIP is more complicated, so we derive the location update and

packet delivery costs, respectively. By the new scheme, if we assume the threshold is K,

as we can see in Fig. 4.2, and if an MH crosses i subnets between two consecutive packet

arrivals, the MH will update to the HA [L ] times and update to the pervious FA in the

rest i ['] times [45]. So the average location update cost function can be written as

+ u K (i-). (4.3)
M'(, K, p) (= (LKtU +(i L- j)S)c(i). (4.3)
K K
i=0

Similarly, we can obtain the packet delivery cost function. In the DHMIP scheme, some

additional cost is introduced. When a packet is tunnelled to the first FA in the MH current

FA hierarchy, if the first FA is not the user's serving FA, then additional FAs have to be









traversed before the packet can reach the destination. In addition to the packet delivery cost

in the Mobile IP scheme, there is additional (i K [f+- K)T cost in the DHMIP scheme.

So the F' can be obtained as follows:


F'(i, K, p)


The a(i) can be express as [48]


a(i)


F+ ~(i+K-[ K)Ta i).
i=0



1 -1-- if i
P
(l-g)2gi 1 if i > 0
P


where g =f, (s), the Laplace transform of the residence time random variable density

function. In order to analyze (4.3) and (4.4), we assume that i = jK + q, then


a(jK + q)


(1 g)2 K)Jqg yzJq
pg


(4.6)


where
(1 g)2 K
y =, z = x = g.
pg
Notice that both 0 < q < K and 0 < K < K, we can rewrite M' as


M'(v, K, p)


S i () + (U S) L i +- (i)
i=0 i=0
a K-1
S y\-C-jK +q +
S+ (U S) [K j(jK + q)
P j=0 q=0
SK-1
-S + (U S)( [ )K + q 1+ yz '
P j=0 q=0
oc K-6-1 K-l+n
S + (U S)y [) Z YX + 0( + 1)Z X]
P j=0 q=0 q=K-n

S + (U S)y(1 x K+) Z .+ (U S)y(xK-n K+n)
p j1-x 1 -x
j=0
S (U- S)y(l- 1K+")z (U S)y(XK- K+)
-+
p (1- x)(1 z)2 (-x)(1- Z)
S (U S)(I g)gK-1 g_ K+n g2n
P + [ -+ ].
p p(1 gK) K


(4.4)





(4.5)


Z 0
j=o0


(4.7)









Similarly, we can obtain


F'(i, K, p)


00 00 00 .
F + iTa(i) + KTa(i) KT L ](i)
i=0 i=0 i=0
T KT(I g)gK-1 gK+K I g 2
F+ + KT -[ -+ ]
p p(1 gK) -K g"


Thus, the total cost is


C'(v, K, p)






Furthermore, if we assume

(4.9) as


M'(, K, p) + F'(, K, p)
S+T
F+--- + KT
P
(U- S- KT)(1 g)gK-1 K+: 1 g2n
-t[ + ]. (4.9)
p(l gK) 1 gK g K

that the K is uniformly distributed with P3 = we can rewrite
Kw a ert


K-1
- C'/
K K
K=0
SS+T (K- )T
F+-+
p 2
(U- S- KT)(I- g)gK
p(l gK)K
SS+T (K- )T
F+--+ +
p 2
K gl-K _
-1 gK I- g


-1 K-1 gK+n 1 2n
0 -l + g]
S=0 gK
(U S KT)(I -g)gK-1
p(I gK)K


(4.10)


For demonstration purposes, we assume that the mobile user subnet residence time is

Gamma distributed with mean 1/[t. The Laplace transform of a Gamma distribution is


fn) = ()- --- )+
8s + w


thus, we have


(4.8)


C uniform(K, p)


g-fT* (A \\ ( 'It )7 ( \^)7
.c/ + fmw P + -)









In particular, when 7 = 1, we have an exponential distribution for the subnet residence time.

If the residence time is exponentially distributed, we have g = then (4.10) is reduced to

S+ T (K- 1)T U -S- KT
C'j (K, p) F + + (+
uniform P 2 (1 + p)K I

(I + P)K (I + P)K 2 -p (4.11)
.[ +) ]+ (4.11)
(1 + p)K- 1 Kp

In reality, many mobile users usually keep accessing the Internet in a vicinity. When

they roam far away from their daily work place, they would access the network less often.

So we can simulate the situations by setting K linearly or exponentially distributed. If K is

linearly distributed with (K= I), then
K(K+1)'


Cna,,,, (K, p)
_0

F+

(U

K-1

0=O

If K is exponentially distributed with /P

K-1
C'exponential (K,p) = /


S+T (K- 1)T
P 3
S KT)(1 g)gK-1
p(l- gK)
2(K c) 1 gK+ I g2
K(K+ ) 1-gK g



-1



+ T e-l+ (K I)e-K-l Ke-K
p (1 -K)(I -1)
S KT)(I g)gK-1
p( gK)

-- ^ ------- \ ---- --- _L ---- ^
-K(1 e-- 1) 1 K+ 1 g2
1-K- 1gK + g
C e-K K 9 K


For MHs with different traffic load and mobility patterns, their optimal hierarchy level

thresholds should be different. The optimal threshold (Kopt) for an MH is the value of K

that minimizes the cost functions derived above. Since the Kopt must be an integer, we

use a similar method in [31, 43] to obtain the optimal values. We define the cost difference


(4.12)


(4.13)









equation between the system with level K and the one with level K 1 (K > 2) as


A(K, p) C'(K, p) C'(K- 1,p). (4.14)

Given the CMR, the algorithm to find the optimal value of K is defined as

Kt1 ifA(2, p) >0
K1 =(4.15)
max{K : A(K,p) <0} otherwise

Notice that the algorithm to get the optimal value is iterative. It is easy to implement;

however it may result in local minima. How to avoid the local minima was discussed in [69].

In practical systems, we can determine the optimal value to a limited predefined maximum

number. Then, the optimal value can be found by evaluating the total costs for each of

the allowed numbers in the range. The estimation of the MH packet arrival rate was also

discussed in [31], which we will use in this work.
4.5 Numerical Results

To evaluate the performance of the DHMIP scheme, we need to know the relative

operation costs. We use the following notations for our analysis.

tmf_ : the transmission cost of location update between an MH and an FA;

mff_ : the transmission cost of location update between FAs;

fh_u : the transmission cost of location update between an FA and the HA;

pfh_ : the location update processing cost between an FA and the HA;

Pmf_. : the location update processing cost between an MH and an FA;

Pffu : the location update processing cost between FAs;

mffd : the transmission cost of packet delivery between FAs;

mhf_d : the transmission cost of packet delivery between the HA and an FA;

mfm-d : the transmission cost of packet delivery between an FA and an MH;

Pff-d : the packet delivery processing cost between FAs;

". _d : the packet delivery processing cost between the HA and an FA;

Pfm_d : the packet delivery processing cost between an FA and an MH.

According to the above definitions and the protocols of the Mobile IP and the DHMIP

schemes described in Section 4.3, the MH registration cost (U) to the HA, the packet delivery

cost (F) in the Mobile IP network, the hierarchy setup cost (S) and the hierarchical packet









Table 4.2: The performance analysis parameters (1)

Pfm_d Pffd l'-'.d mfmd mffd
1 0.5 10 5 25
Pmfu Pff-u Pfhu mrmfu ffu
2 1 20 10 50

Table 4.3: The performance analysis parameters (2)

set mfh u mhf_d
1 250 125
2 500 250
3 1000 500


delivery cost (T) in the DHMIP can be expressed as:


U = 2mmfu + 2mfhu + 2pmfu + Pfh_; (4.16)

S = 2mmf_u +2mffu+ 2pmf_ +ffu; (4.17)

T = mffd+pffd; (4.18)

F = mhfd +mmfmd +-ld +Pfnmd. (4.19)


The Mobile IP scheme may result in heavy system traffic load because of the MHs'

location update messages. The DHMIP scheme attempts to reduce the registrations to the

HAs by informing MHs' new location to the old FAs. This would effectively reduce the

long distance signaling traffic at the expense of increasing the local traffic load. Usually,

an MH's new FA is close to its old one in term of the hops between them and the traffic

load is distributed evenly among the system, so that the network can accept more service

requests. It is obvious that the DHMIP scheme performance depends on the relative cost of

the long distance registration to the HA to the local registration to FAs. In this chapter, we

use the parameters in Table 4.2 and Table 4.3 for performance analysis. We use three sets

of parameters in our analysis. The signaling costs between the .,li.1:ent hierarchy levels are

fixed and the costs between the HA and an FA are 5, 10 and 20 times of those in the three

sets of parameters, respectively.

Fig. 4.3 shows the total signaling costs for Mobile IP and the DHMIP schemes under

various CMR values. In this figure, we observe that when the CMR is small, the DHMIP

scheme generates much less traffic than the Mobile IP scheme does even without using the










MIP
DHMIP with K=4
DHMIP with Kpt

104



0
-\









10




101 100 101 102
Call-to-Mobility Ratio (p)

Figure 4.3: Comparison of the total costs for different schemes


optimal values. We also see that our scheme can reduce the traffic load even further with

the optimal values. How to obtain the optimal value has been discussed in Section 4.4. If we

examine Fig. 4.3 more carefully, we can see that the total cost for the DHMIP scheme with

fixed threshold exceeds that for the Mobile IP scheme when the CMR is large. This can

be seen more clearly in Fig. 4.4. In this figure, we show the relative costs for the DHMIP

scheme to that for Mobile IP scheme under different CMR. The uniform distribution is

used and the thresholds are four for all the curves in this figure. It is obvious that the total

cost for the DHMIP scheme with fixed threshold can exceed that for the Mobile IP scheme

when the CMR is large. In the new scheme, the packets for an MH will be processed and

tunnelled through more FAs before it reaches the destination. When the packet arrival rate

is high comparing with the mobility and the threshold is fixed, the total cost may exceed

that for the Mobile IP scheme. However, we can show that, with the optimal values, the

DHMIP scheme will never generate more traffic than the Mobile IP scheme does under any

conditions. We can also see in Fig. 4.4 that the scheme performs better when the relative

cost to the HA is higher, the reason is straightforward.

Some mobile users usually roam only in a vicinity when they access the internet ap-

plications. For example, a passenger waiting for his/her flight in terminal or the people in

some construction field. They usually require less service when they move away from some







73

1.3

12Set 3
1.2 -2
Set 2



1./ /
Set-





S0.8-

0.7-

0.6 -

0.5 --



0.3
10 100 101 102
Call-to-Mobility Ratio (p)

Figure 4.4: Comparison of the total costs for uniformly distributed c under different param-
eters


specific areas. In this chapter, we use the linear and exponential distributions to simulate

those kinds of mobile users and show the performance in Fig. 4.5. It can be seen that, with

fixed threshold, the users roaming around a specific area can generate less signaling traffic

under the DHMIP scheme. In fact, when a mobile user roams in a specific area, there is high

probability that the user will revisit some subnets. This will generate loops in the hierarchy

and result in signaling traffic which can be removed by "loop removal" introduced in Section

4.7.

Before discuss the choice of the optimal threshold for a mobile user, we study the effect

of the hierarchy level number on the DHMIP scheme performance first. Under various CMR

values, we change the number of hierarchy levels K in Fig. 4.6. When the CMR is low,

meaning that the user has relatively high mobility, larger K can minimize the total signaling

costs because more updates to the HA can be replaced by hierarchy setup, which will cost

less. However, when the CMR is high, the packets delivered to an MH have to travel more

FAs before reaching the destination with larger K. In this case, the total costs of the DHMIP

scheme may exceed those of the MIP scheme. We can also see in Fig. 4.6 that it is hard to

find an optimal fixed threshold for an MH under different situations. In order to achieve the

best performance, the selection of threshold must be dynamic.





















uniform
linear
exponential


Call-to-Mobility Ratio (p)



Figure 4.5: Comparison of the total costs under different K distributions


1.2


1.1 K- 2K=
K=4

1-


0.9


0.8 ---:
.o0


//
0.7 -




/- /
0.6 -


0.5-


0.4-


0.3


0.2
10 100 101 10:
Call-to-Mobility Ratio (p)



Figure 4.6: Comparison of the total costs under different K



















uniform
linear
exponential


Call-to-Mobility Ratio (p)


Figure 4.7: Comparison of the total costs under different K distributions with optimal K


01
-1 -0.5 0 0.5 1 1.5 2
Call-to-Mobility Ratio Iglo(p)


Figure 4.8: The optimal number of hierarchy levels for uniform distribution







76

1.1 1.1 1
y=00.1 y=00.1 y=00.1
y=1 y=1 0=1
1 y=100 1 y=100 0.9 y=100

0.9 0.9 0.8

S0.8 0.8 0.7

0.7 0.7 0.6

0.6 0.6- 0.5

0.5.5 0.5 0.4
~04 0
0.4- 0 0.4 0.3-

0.3 / 0.3 0.2-

0.2- 0.2- / 0.1

0.1 0.1 I 0
10 10 10 2 1 1010 1 10 2 1 1010 1 10 102
P P P

Figure 4.9: The effect of the variance of the FA residence time (7)


In Fig. 4.7, we show the relative performance for the DHMIP scheme using the optimal

values under different CAMR values and different K distributions. We can observe that

our new scheme can reduce 85% of the signaling cost when the CAMR is small. With the

increase of CAMR, the relative cost also increases. However, the cost for the DHMIP scheme

will never exceed the cost for the MIP scheme. Fig. 4.8 shows the optimal values under

different CAIR for the three sets of parameter. In this figure, we assume that the K for the

MH follows a uniform distribution. In Fig. 4.8, the optimal values decrease with the increase

of CMAR. When the relative update cost to the HA is large, an MH can have relatively large

thresholds. The reason is obvious. If the update cost to the HA is high, an MH can generate

less signaling traffic by setting up more FA hierarchy, otherwise, the MH should update the

HA more often, which implies smaller K. In fact, when the optimal value equals one, our

new scheme becomes the MIP scheme, so that the total costs will never exceed those for

the MIP scheme. Based on the above analysis, we can see that the MIP scheme is suitable

for mobile users with high CMR and our DHMIP scheme can perform well for all kinds of

users.

We next investigate the sensitivity of the performance costs and the benefits of the

DHMIP scheme to the variance of the mobile user's mobility pattern. In our analysis, we

assume that the packet arrivals to a mobile user form a Poisson process and the user's






77

residence time in a subnet has a Gamma distribution. For a Gamma distribution, the

variance is V = That is, a large 7 implies a small variance. Fig. 4.9 demonstrates the

effects of the variance of the FA residence time under three distributions. In this figure, we

assume 7 = 0.01, 1 and 100, respectively. We can see in the figure that, the variance of the

FA residence time in the subnets does not affect the performance of the DHMIP scheme very

much. We derive the optimal values based on the exponential residence time distribution.

When the 7 = 0.01 or 100, those values are not optimal anymore. However, the DHMIP

can still achieve similar performance under different CMR, different residence time and K

distributions. This phenomenon proves the flexibility and the adaptivity of our new scheme.

4.6 Comparison with the IETF Hierarchical Scheme

The IETF Hierarchical Mobile IP (HMIP) protocol aims to reduce the signaling traffic

to the home networks while reducing the signaling delay. The specification of the HMIP

protocol can be found in [42]. In this protocol, the HMIP aini'l-,v- a hierarchy of FAs to

locally handle Mobile IP registration. When an MH moves from one subnet to another, it

performs a home registration to its HA. During a home registration, the HA registers the

care-of address of the MH. When the visited domain supports regional tunnel management,

the care-of address that is registered at the HA is the publicly routable address of a Gateway

Foreign Agent (GFA). This care-of address will not change when the mobile node changes

foreign agent under the same GFA. In this situation, the MH will perform a regional update

to GFA, the MH mobility management can be handled locally in this way. During the

communication, when packets are sent to the MH, the user's HA intercepts the packets,

encapsulates them and tunnels them to the CoA of the MH. Those packets will reach the

GFA of the MH through the network. The GFA checks its visitor list and re-tunnels the

packets to the user's corresponding FA. The FA further relays the packets to the MH.

Typically, one level of hierarchy, where all FAs are connected to the GFAs, is considered;

however the protocol may be utilized to support multiple levels of hierarchy as discussed in

Appendix B in [42].

In this section, we compare our DHMIP scheme with the IETF hierarchical mobile IP

scheme. We assume U1 is the MH location update cost to a GFA, U2 is that to the HA and

Fgf is the additional packet delivery cost induced by the GFA. In the IETF HMIP scheme,









the system structure is fixed. If an MH roams under the same GFA, the user total mobility

management costs can be reduced. However, the HMIP protocol requires the MH to perform

home registration when user enters a subnet charged by another GFA. In order to analyze

the HMIP performance, we need to know the probability that a user moves out of a regional

network. We further assume that cc is the probability that the next movement of MH is

under the same GFA. In order to obtain c, we assume that there are N subnets in a foreign

network and m subnets in one regional area, respectively. In the model, we define the action

that each MH moving out of a subnet a movement. In one movement, an MH can visit a

subnet randomly. Then, we can have c = '-1. Usually, the distance between a GFA and

the MH's current FA is farther than the FAs in our DHMIP scheme in term of hops, so

the location update and packet delivery costs are also higher. Conservatively, we define the

cost coefficient ( = 2. The GFA processing cost is proportional to the number of subnets

under it. Since IP routing table lookup is based on the longest /". Ipi matching, then the

GFA processing complexity is proportional to the logarithm of m. According to the above

assumptions we can have U1 = (Slog(m), U2 U U1 and Ff = (Tlog(m). Thus, the

total cost function for IETF HMIP scheme can be expressed as
00
C'HMIP () i[wU + (-I )U2 i) + F + Ff
i= 0
SU1 + (1 +c)U2
+F+ F,

S+ F + Tlog(m). (4.20)
p(N 1) p

In Fig. 4.10, we plot the relative cost curves for the IETF HMIP scheme and DHMIP

scheme to that of the basic MIP protocol. In this figure, we assume N = 50, m = 10 and

20, respectively. We can see in this figure that our DHMIP scheme generates less system

signaling cost than the IETF HMIP scheme under all the conditions even without using the

optimal threshold values. We can also see that if the number of subnets under one GFA

is large, the HMIP scheme can reduce the system signaling traffic more when the CMR is

small. However, the total cost increases rapidly with the increase of CMR and exceeds that

with smaller m. So, for the IETF HMIP scheme, it is hard to find an optimal number of

subnets under one GFA for all the users in one network. Another drawback of the HMIP







79

1.2 -






0.8


0 0.6-


H-
0.4


0.2 -- DHMIP
=19/49
=9/49
Non optimal
101 10 101 102
Call-to-Mobility Ratio (p)

Figure 4.10: The comparison with the IETF hierarchical scheme


scheme, as mentioned in [43], is that the centralized system architecture makes the system

performance sensitive to the failure of GFAs. Our DHMIP scheme avoids this problem

successfully.

4.7 Improvement of DHMIP Scheme

The performance of the DHMIP scheme can be improved further by the state activation

and loop removal.

4.7.1 State Activation

Currently, Mobile IP supports registrations but not paging. IP paging is an IP layer

process for an IP network to determine an MH's precise location, i.e., the precise attachment

point to the network where it can receive IP packets. With the IP paging support, MHs

can reduce the registration to the HA when they do not engage any communication. In

the IP paging protocol, the HA only keeps approximate location information of MH which

is in idle state. The MH will turn to active state when packets are received and update

its current location to the HA. We can enhance the performance of the DHMIP scheme by

setting up a long threshold for MH in idle state in a similar way. In this work, we name

the improved DHMIP scheme Enhanced DHMIP scheme. Similar to the Cellular IP [41], we

define the idle MH as one that has not received data packet for a system specific period.

In the Enhanced DHMIP scheme, the MH can keep a fixed relatively large threshold K so









that the MH can avoid updating the home network often. An idle mobile host that receives

the first packet moves from idle to active state and updates its current location to the HA

immediately and keeps doing it when it enters new subnets if it keeps in active state. In the

Enhanced DHMIP scheme, the data packets can avoid the extra transmission delay in the

DHMIP scheme and the total signaling traffic is reduced at the same time. If the mobile

user has not received packet for some predefined period, the MH returns back to the idle

state again. In this enhanced version, an MH can maintain a fixed threshold K without

computing and changing it from time to time. This can also reduce the mobile terminal

energy consumption. For different mobile users with different call to mobility patterns, they

can set different predefined threshold values. In our analysis, we assume that the MH turns

to active state after receiving the first data packet and can update to its home network by

attaching the location information to the acknowledge messages. Then, we can obtain the

location update cost function (EM'), packet delivery cost function (EF') and the total cost

function (EC') as follows:


EM'(K, p)




EF'(K, p)




EC'(K, p)


OO
[L KU+^ (-^ [K)S a(i),
i=0
S (I g)gK-1(U S)
-+
S p(l gK)
oO
F +j(i- K[-)Ta(i)
i=0
S-[1- KgK-1 (K- 1)gK]T
F +
p(l gK)
EM'(K, p) + EF'(K, p)
SS (U- S)(I g)gK-1 + [I KgK-1 + (K- I)gK]T
P p(l gK)
SS+T (U- S- KT)( g)gK-1
F P p(
p p(l- gK)


(4.21)




(4.22)






(4.23)


We further assume that the MH's FA residence time in subnets is exponentially distributed,

then g = we can rewrite (4.23) as

T+S U-S-KT
EC' (K, p) = -F (4.24)
p (1 + p)K 1












1I
0.9- -K=2
K=4
K=10
0.8 :

0.7 -
b/
S/ /
0.6 -


S /
0.5 -i

0.4 -

0.3 -

0.2

0.1
10 100 101 102
Call-to-Mobility Ratio (p)


Figure 4.11: The total cost for the enhanced DHMIP scheme


Fig. 4.11 shows the performance of the Enhanced DHMIP scheme. We can observe

from the figure that there is no optimal threshold for an MH. The larger the threshold, the

better the performance. However, a large threshold may generate long packet delivery delay

for the first data packet so that the network operator should choose the value by considering

the signaling saving and the cil.,lil'r of services comprehensively. In Fig. 4.11, we also see

that the K has little effect on the scheme performance when the CMR is large. The reason

is that the MH is always in active state and keeps updating its newly obtained CoA to the

HA in this situation and the effective threshold reduces to one no matter what the predefined

value is. In the Enhanced DHMIP scheme, the total system signaling cost will never exceed

that of the MIP scheme under all the conditions.

4.7.2 Loop Removal

The performance for the DHMIP scheme can also be improved by removing the loops

formed during the mobile user's roaming. In reality, many mobile users roam in a limited

region, for example, in an office building. If the mobile users revisit some subnets they

have visited before, loops may form in the hierarchical architecture in the DHMIP scheme

according to the protocol described before. With the loop removal, the MHs can update their

new care-of addresses less often and the total signaling cost can be minimized further [23].

When an MH enters a new subnet and tries to set up a new level of hierarchy, the new FA








checks its hierarchy list first. If the FA is already in the MH hierarchy, the FA can delete

the subsequent FA addresses for that user without updating the old FA so that the loop is

removed. In this section, we analyze the loop removal effect on the DHMIP performance.

If we consider the MH revisiting scenario, we can define 0(i) as the effective number

of subnets crossed between two consecutive packet arrivals. Although the loop removal can

reduce the number of HA updates, the MH has to update its location to the current FA even

the MH has visited that FA before. We define 6 = 2mrmfu + 2Pmf_, as the signaling cost

of an MH notifying its arrival to the current FA, we can obtain the cost function with loop

removal as

K-_l o (i)+ 0 (i)
NC'(,K,p) [(U 6) (0(i)- L K )(S-6)
KO i=0

is + (0(i) + L (i)+ )+ F
KP K-1
F F+ +T 7&c+(U-S-KT)>[>(i)Kja(i)
SK=O K=O i=0

+(S + T- 6) 0(i)(i). (4.25)
i=0
For simplicity and demonstration purpose, we assume that K is uniformly distributed and

0(i) = i, where C is the percentage of the effective number of subnets a user visited to the

total number of subnets the user visited. Then, we have

(K I)T (S + T) + (1 C)6
NC'(K, p) = F+ +
2 p
(U S KT) K-Q I$ i + K (.26)
+ KLK (i). (4.26)
K=O i=0


Fig. 4.12 shows the loop removal effect when C is 1, 0.8, 0.5 and 0.2, respectively. When

C = 1, it is the worst case that the mobile user never revisits any subnet he/her visited
before. From Fig. 4.12, we can see that the total signaling cost for the DHMIP scheme can

be reduced by the loop removal mechanism.
4.8 Conclusions

This chapter has presented a new location management scheme for Mobile IP network

the Dynamic Hierarchy Mobile IP (DHMIP) strategy. Instead of updating the home networks







83



12


=0 5
=0 2


08-


0 0 6 -0
S06


04-


02

0

10 100 10' 102
Call-to-Mobility Ratio (p)

Figure 4.12: The total cost with loop removal


far away, in the DHMIP scheme, the MHs inform their new care-of addresses to the previous

FAs. When data packets arrive at the MHs, the packets can be delivered through the

FA hierarchy. The transmission distance between the FAs is usually shorter than that

between an MH and the HA, so the total signaling cost can be reduced. In this chapter,

we also proposed an iterative algorithm to compute the optimal threshold values for specific

users with different call to mobility patterns. Our analysis proves that the DHMIP scheme

outperforms the traditional MIP scheme under various conditions. In order to improve the

MIP performance, the Hierarchical Mobile IP (HMIP) protocol has been proposed by IETF.

In the IETF HMIP scheme, the network architecture is centralized and sensitive to the failure

of GFA. It is also proved in our analysis that it is impossible to find an optimal number

of subnets under one GFA for all kinds of users. We compared the IETF HMIP scheme

with our DHMIP in this chapter, the results show that the DHMIP scheme can achieve

better performance. The DHMIP performance can also be improved by considering the

state activation and it loop removal. All the analysis shows that the DHMIP can minimize

the location management cost for Mobile IP networks and has more flexibility and robustness

than the traditional Mobile IP scheme.














CHAPTER 5
CONCLUSIONS AND FUTURE WORKS


The rapid growth of wireless mobile networks and services, fuelled by the next generation

mobile communications systems research, has ushered in the era of ubiquitous computing.

Lightweight portable computers, IP-based appliances, and the popularity of the Internet

are providing strong incentives to service providers to support seamless user mobility. It is

expected that future wireless mobile networks will be more heterogeneous and that every

mobile user will be able to gain access to the Internet backbone by attaching his or her mobile

computing devices to a wireless access point. Besides some new technologies required to make

the above scenarios realized, the user mobility management protocols are confronted with

more and more challenges.

Mobility is an essential characteristic of the wireless communication networks. With the

increase of the mobile user number and the decrease of the registration area size, the location

management traffic will increase dramatically in the 3rd generation wireless communication

signaling networks. The research aiming to minimize and optimize the mobility management

cost has been carried out extensively. In this dissertation, we proposed two schemes based

on pointer forwarding and one based on user mobility pattern for wireless communication

networks. The pointer forwarding schemes include two-level pointer forwarding and POFLA

schemes. In both schemes, we try to improve the original mobility management schemes by

introducing a new functional entity-Mobility Agent (\1.A). The MAs can either be software-

based or hardware-based and can be distributed dynamically among the network according

to the users' current traffic and mobility. This characteristic makes the new schemes more

adaptive and robust. In both schemes, we designed two kinds of pointers for different charge

domains: high level pointer is set up when the threshold for low level pointer chain is reached

and home location registration is performed after the high level pointer limit is reached. The

two new schemes improve the network mobility management performance greatly comparing

to some other proposed schemes, such as per-user forwarding and location anchor strategies.

84









The two-level pointer forwarding and POFLA schemes avoid the bottleneck effect in the

anchors for the local anchor scheme and overcome the small pointer length limit for the

per-user forwarding strategy. In our works, we developed analytical models for the wireless

communication systems, analyzed and obtained the cost functions for the aforementioned

protocols. The results give a clear picture of the relative performance for all the schemes.

Our new pointer forwarding schemes outperform the original protocols and some improved

versions under various conditions. Based on the observation that many mobile users follow

some fixed daily routines, we proposed another new mobility management scheme-user

mobility patter based scheme (\11PBS). In this scheme, the users' mobility patterns are

stored in the user profiles and hence the network can determine the users' current location.

The MPBS scheme cannot only reduce the system location management signaling traffic but

also make it possible for the service providers to provide users user-oriented services. The

system resources can be allocated more reasonably and efficiently because the user mobility

pattern is included in the user profile and accurate prediction can be made by the system

according to the information. We verified our results by simulations in our works. All

the results -I.'-.I -1 that the MPBS scheme can significantly improve the network mobility

management performance. The total cost for MPBS scheme increases slowly with the unit

paging cost. It is an impressive advantage for the 3rd generation systems because the radio

transmission cost will keep increasing in the future.

The 3rd Generation Partnership Project (3GPP) and 3rd Generation Partnership Project

2 (3GPP2) are the consortium of international standards bodies tasked with developing ar-

chitectures and standards for the third-generation cellular systems. Although more harmo-

nization is needed before the final standards be accepted by all the nations, it is now widely

recognized that using IP as the foundation for next-generation mobile networks makes strong

economic and technical sense. Traditionally, wide-area mobility over the IP networks has

been based on the family of Mobile IP protocols. However, the micro-mobility of the Mobile

IP performance needs to be enhanced because of the heavy signaling traffic and long latency.

In our research, we extended the IETF regional registration protocol by proposing the Dy-

namic Hierarchy Mobile IP location management scheme (DHMIP). In the DHMIP scheme,

an MH updates its new location to the previous FA so that a dynamic hierarchy is set up






86

for the specific user. Since the hierarchies are distributed among the network and adjusted

based on the users' up-to-date traffic burden and mobility, the new scheme can improve the

IP network mobility management performance greatly. The new DHMIP strategy also suc-

cessfully avoids the single failure point problem in the IETF regional registration protocol

so that the network robustness and survivability are enhanced at the same time. The most

important contribution of our research is the analytical approach we developed in this work.

Almost all the previous works dealing with the IP network mobility were carried by simula-

tions. In our research, we developed a systematic model and derived the close-form solution

for the cost functions for the Mobile IP and DHMIP schemes. It is more straightforward to

find out how the network configuration parameters can affect the protocol performance. We

hope the new approach we developed can open a new avenue to the future research in this

area and the model can be consummated by other researchers.

In our research, when we develop the system model and analyze the performances, we

always assume the Poisson and exponential distributions for some variables. The reason

we use the above distributions is that they are easy to analyze so that we can obtain an-

alytical expressions. In reality, those assumptions may not be realistic. How to generalize

our analytical results under more realistic assumption becomes an important research task.

Another issue about the DHMIP protocol is the user data arrival pattern. With more and

more kinds of services provided over the Internet, especially the real time data services, it

is becoming more and more difficult to model the packet transmissions. In our model, if the

packet arrival can be replaced by the service session, the results could be more persuasive.

However, it is still an open issue about the session model over the multimedia networks

based on IP protocol.














REFERENCES


[1] I.F. Akyildiz, J. McNair, J.S.M. Ho, H. Uzunalioglu and W. Wang, -\lMbility Manage-
ment in Next-Generation Wireless Systems," Proc. of the IEEE, vol.4, no.5, pp.1347-
1384, October 1996.

[2] V.W.S. Wong and V. C. M. Leung, "Location Management for Next-Generation Per-
sonal Communications Networks," IEEE Network, vol.14, no.5, pp.18-24, Septem-
ber/October 2000.

[3] Theodore S. Rappaport, A. Annamalai, R. M. Buehrer and William H. Tranter, "Wire-
less Communications: Past Events and A Future Perspective," IEEE Communications
Magazine, vol.40, no.5, pp.148-161, M\.ir, 2002.

[4] S. Tabbane, "Location Management Methods for Third-Generation Mobile System,"
IEEE Communication Magazine, vol.35, no.8, pp.72-84, August, 1997.

[5] D. C. Cox, \\ i. i -- Personal Communications: What Is It?" IEEE Personal Com-
munications, vol.2, no.2, pp.20-35, April, 1995.

[6] Y. B. Lin, "Determining the User Locations for Personal Communications Services
Networks," IEEE Trans. Veh. T-,-;,.,.l''i i. vol.43, no.3, pp.466-473, Aug. 1994.

[7] EIT/TIA, "Cellular Radio Telecommunications Intersystem Operations," Technical
Report IS-41 (Revision B), EIA/TIA (1991).

[8] ETSI, "Digital Cellular Telecommunications System (Phase 2+): Mobile Application
Part (\I!AP) Specification (CSMI 09.02 version 7.51 Release)," 1998.

[9] Y.B. Lin, Y.R. Haung, Y.K. Chen and I. Chlamtac, "Mobility Management: from
GPRS to UMTS," Wireless Communications and Mobile Coi,,,tuin lu (WCMC), vol.1,
no.4, pp.339-359, 2001.

[10] Y. B. Lin, "Reducing Location Update Cost in a PCS Network," IEEE/AC-[ Trans.
on Networking, vol.5, no.l, pp.25-33, Feb. 1997.

[11] Charles E. Perkins, '-\l.bile IP," IEEE Communications Magazine, vol.5, no.2, pp.84-
99, Mi.,v, 1997.

[12] S. Das, A. Misra and P. Agrawal, "TeleMIP: Telecommunications-Enhanced Mobile IP
Architecture for Fast Intradomain Mobility," IEEE Personal communications, vol.7,
no.4, pp.50-58, August, 2000.

[13] Andrew T. Campbell, Javier Gomez, Sanghyo Kim, Chieh-Yin Wan, Zoltan R. Turanyi
and Andras G. Valko, "Comparison of IP Micromobility Protocols," IEEE Wireless
Communications, vol.9, no.l, pp.72-82, Feburary, 2002.

[14] I. F. Akyildiz and J. S. M. Ho, "On Location Management for Personal Communica-
tions Networks," IEEE Comm. Mag., vol.34, pp.138-145, Sept. 1996.
87









[15] W. Wang, I. F. Akyildiz, "A New Signaling Protocol for Intersystem Roaming in Next-
Generation Wireless Systems," IEEE Journal on Selected Areas in Comm., vol.19,
no.10, pp.2040-2052, Oct. 2001.

[16] C. Perkins, "IP Mobility Support," Internet RFC 2002, October 1996. Work in
progress.

[17] S. Mohan and R. Jain, "Two User Location Strategies for Personal Communications
Services," IEEE Personal Communications, vol.1, no.l, pp.42-50, 1994.

[18] I. F. Akyildiz and J. S. M. Ho, "On Location Management for Personal Communica-
tions Networks," IEEE Communications Magazine, vol.34, no.9, pp.138-145, Septem-
ber, 1996.

[19] J. S. M. Ho, I. F. Akyildiz, "Dynamic Hierarchical Database Architecture for Location
Management in PCS Networks," IEEE/ACM Trans. Networking, vol.5, no.5, pp. 646-
661, Oct. 1997.

[20] R. Jain, Y.B. Lin, C. Lo, and S. Mohan "A Caching Strategy to Reduce Network
Impacts of PCS," IEEE Journal on Selected Areas in Comm., vol.12, no.8, pp.1434-
1444, Oct. 1994.

[21] N. Shivakumar and J. Widom, "A New Signaling Protocol for Intersystem Roaming
in Next-Generation Wireless Systems," IEEE Journal on Selected Areas in Comm.,
vol.19, no.10, pp.2040-2052, Oct. 2001.

[22] J. Ho and F. Akyildiz, "Local Anchor Scheme for Reducing Signaling Costs in Personal
Communications Networks," IEEE/ACM Trans. on Networking, vol.4, no.5, pp.709-
725, October 1996.

[23] R. Jain and Y.B. Lin, "An Auxiliary User Location Strategy Employing Forwarding
Pointers to Reduce Network Impacts of PCS," Wireless Networks 1, pp.197-210, 1995.

[24] J. Z. Wang, "A Fully Distribute Location Registration Strategy for Universal Personal
Communication Systems," IEEE Journal on Selected Areas in Comm., vol.11, no.6
pp.850-860, Aug. 1993.

[25] B. R. Badrinath, T. Imielinski and A. Virmani, "Locating Strategies for Personal
Communication Network," Proc. Workshop Networking of Personal Communications
Applications, Dec. 1992.

[26] V. Anantharam, M. Honig, U. Madhow, and V. Wei, "Optimization of a Database
Hierarchy for Mobility Tracking in a Personal Communications Network," Performance
Evaluation, Vol. 20, pp. 287-300, 1994.

[27] H. W. Hwang, M. F. Chang and C. C. Tseng, "A Direction-Based Location Update
Scheme with A Line-Paging Strategy for PCS Networks," IEEE Communications Let-
ters, vol.4, no.5, M..ir, 2000.

[28] J. Li, H. Kameda and K. Li, "Optimal Dynamic Mobility Management for PCS Net-
works," IEEE/ACM Trans. on Networking, vol.8, no.3, pp.319-327, June, 2000.









[29] G. Krishnamurthi, M. Azizoglu and A. K. Somani, "Optimal Distributed Location
Management in Mobile Networks," Mobile Networks and Applications, vol.6, pp.117-
124, 2001.

[30] I. F. Akyildiz and J. S. M. Ho, "Dynamic Mobile User Location Update for Wireless
PCS Networks," ACM-Baltzer J. Wireless Networks, vol.1, no.2, pp.187-196, July
1995.

[31] H. Xie, S. Tabbane and D. Goodman, "Dynamic Location Area Management and
Performance Analysis," Proc. IEEE VTC'93, pp. 536-539, Secausus, NJ, USA.

[32] A. Bar-Noy, I. Kessler and M. Sidi, -\l.bile User: To Update or Not to Update,"
ACM-Baltzer J. Wireless Networks, vol.1, no.2, pp.175-185, July. 1995.

[33] I. F. Akyildiz, J. S. M. Ho and Y. B. Lin, "Movement-based Location Update and Se-
lective Paging for PCS Networks," IEEE/ACM Trans. Networking, vol.4, no.4, pp.629-
636, Aug. 1996.

[34] V. W. S. Wong and V. C. M. Leung, "An Adaptive Distance-Based Location Update
Algorithm for Next-Generation PCS Networks," IEEE Journal on Selected Areas in
Comm., vol.19, no.10, pp.1942-1952, Oct. 2001.

[35] C. Rose, \lilii :iin the Average Cost of Paging and Registration: A Timer-based
Method," AVM-Baltzer J. Wireless Networks, vol.2, no.2, pp.109-116, 1996.

[36] L. P. Araujo and J. R. Boisson de Marca, "Paging and Location Update Algorirhms for
Cellular Systems," IEEE Tran. on vehicular technology, vol.49, no.5, pp.1606-1614,
2000.

[37] C. Rose and R. Yates, *liii :ing the Average Cost of Paging Under Delay Con-
straints," ACM-Baltzer J. Wireless Networks., vol.1, no.2, pp.490-495, July. 1995.

[38] U. Madhow, M. L. Hoing and K. Steiglitz, "Optimization of Wireless Resource for
Personal Communications Mobility Ti.. in.-." IEEE INFOCOM'94, pp.577-584.

[39] P. J. McCann and I. Hiller, "An Internet Infrastructure for Cellular CDMA Networks
Using Mobile IP," IEEE Personal Communications, vol.7, no.4, pp.26-32, Aug. 2000.

[40] Ramachandran Ramjee, Kannan Varadhan, Luca Salgarelli, Sandra R. Thuel, Shie-
Yuan Wang and Thomas La Porta, "HAWAII: A Domain-Based Approcah for Sup-
porting Mboility in Wide-Area Wireless Networks," IEEE/ACM Trans. of Networking,
vol.10, no.3, pp.396-410, June 2002.

[41] A. T. Campbell, and J. Comez, "An Overview of Cellular IP," IEEE WCNC'99, New
Orleans, pp. 606-610, Sept. 1999.

[42] Eva Gustafsson, Annika Jonsson and Charles Perkins i\lbile IP Regional Regis-
tration," Internet draft, draft-ietf-mobileip-reg-tunnel-O 4.txt, work in progress, March
2001.

[43] Jiang Xie and Ian F. Akyildiz, "A Distributed Dynamic Regional Locaion Management
Scheme for Mobile IP," Proc. IEEE INFOCOM 2002, pp. 1069-1078, 2002.