Doctoral Thesis ProposalSystems Optimization in Mobility Management

Ashutosh DuttaDepartment of Electrical Engineering

Columbia Universityadutta@research.telcordia.com

May 8, 2006

i

Abstract

Mobile networking technology seeks to preserve the user experience as the user moves andthereby crosses from one network into another. A mobility event is the result of one network con-nection path being replaced by another via the rebinding of common system properties that mayhappen at any one of several protocol layers of the mobile. The rebinding is a sequential processand each process takes finite amount of time. This overall process results in generating a periodof time in which network service is degraded by transient data loss and increased end-to-end delay.Formal techniques to characterize this problem and to develop optimization methodologies for theseprocesses have never been studied as part of a systematic framework for mobility solutions. Thisdissertation develops a systematic systems model that analyzes the systems properties associatedwith a mobility event and characterizes several systems optimization techniques for these processes.

The proposed formal mobility systems model represents these common properties in terms of aset of communicating finite state machines. I analyze this general mobility systems framework anddevelop several methodologies that can model the systems optimization techniques. In particular,I develop methodologies to model the optimization for many of these properties such as discovery,configuration, binding update, media rerouting and apply these methodologies to prototype mobilitysystems that can support subnet, domain and inter-technology roaming for realtime and streamingapplication in the wireless Internet. Some of these methodologies include proactive network andresource discovery, pre-authentication, pre-configuration, reduction of binding delay and minimizingthe effect of media redirection delay by means of dynamic buffering and multicasting.

I validate the mobility system model by using it to analytically assess related mobility protocols.The model will result in an analytical assessment of the techniques that can be directly comparedto our experimental results taking into account several parameters such as mobility rate, packet-to-mobility ratio, simultaneous mobility, distance between communicating nodes and network band-width. I perform a comparative analysis of our optimized mobility management schemes with thesimilar network layer mobility protocols and other fast-handoff mechanism.

ii

Contents

1 Introduction 1

2 Related Work 22.1 Mobility protocols for interactive traffic . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.1.1 Mobility optimization for interactive traffic . . . . . . . . . . . . . . . . . . . 32.2 Mobility protocols for multicast streaming . . . . . . . . . . . . . . . . . . . . . . . 3

2.2.1 Mobility optimization for multicast streaming . . . . . . . . . . . . . . . . . 42.3 Systems model for mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 Mobility Management Introduction 5

4 Handoff Performance Metrics 6

5 Systems Analysis of Mobility Event 8

6 Systems Model for Mobility Event 116.1 Layered approach to mobility optimization . . . . . . . . . . . . . . . . . . . . . . . 116.2 Petri net model for mobility event . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7 Optimization Methodologies and Experimental Validation 157.1 Reactive methodologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

7.1.1 Maintain direct media path between CH and MH . . . . . . . . . . . . . . . 167.1.2 Minimize layer 3 configuration process . . . . . . . . . . . . . . . . . . . . . 187.1.3 Limit binding update and redirect media . . . . . . . . . . . . . . . . . . . . 197.1.4 Maintain security binding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

7.2 Proactive methodologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227.2.1 Make-before-break technique . . . . . . . . . . . . . . . . . . . . . . . . . . 227.2.2 Media independent proactive handover . . . . . . . . . . . . . . . . . . . . . 23

7.3 Multi-layer methodologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257.4 Methodologies for simultaneous mobility . . . . . . . . . . . . . . . . . . . . . . . . 267.5 Methodologies for multicast streaming . . . . . . . . . . . . . . . . . . . . . . . . . 27

8 Roadmap for Future Work 298.1 Validation of mobility optimization methodologies using Petri net models . . . . . . 298.2 Analytic comparison of SIP-based mobility and MIPv6 . . . . . . . . . . . . . . . . 298.3 Comparison of MPA with other fast-handoff mechanism such as FMIPv6 . . . . . . 308.4 Analysis of ping-pong effect during proactive handover . . . . . . . . . . . . . . . . 308.5 Summary of plan for completion of research . . . . . . . . . . . . . . . . . . . . . . 30

9 Conclusion 30

iii

List of Figures

1 Effect of handoff delay due to mobility event . . . . . . . . . . . . . . . . . . . . . . . 62 Infrastructure-based mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Handoff latency components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 Finite state machine model L2 transition . . . . . . . . . . . . . . . . . . . . . . . . . . 145 A generalized Petri net model for mobility event . . . . . . . . . . . . . . . . . . . . . . 146 Petri net model with layered events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Experimental mobile multimedia testbed . . . . . . . . . . . . . . . . . . . . . . . . . . 168 End-to-end delay improvement using direct path . . . . . . . . . . . . . . . . . . . . . . 179 Round trip delay comparison - MIP-LR vs. Mobile IP . . . . . . . . . . . . . . . . . . . 1810 Mobility optimization with media redirection . . . . . . . . . . . . . . . . . . . . . . . 2111 Effect of security rebinding and its optimization . . . . . . . . . . . . . . . . . . . . . . 2212 Multilayer mobility management scenario and results . . . . . . . . . . . . . . . . . . . 2613 Probability for handoff with simultaneous mobility . . . . . . . . . . . . . . . . . . . . 2714 Proxy-based fast-handoff for multicast streaming . . . . . . . . . . . . . . . . . . . . . 2915 Join latency and its optimization in 802.11 environment . . . . . . . . . . . . . . . . . . 29

List of Tables

1 Survey of mobility management protocols . . . . . . . . . . . . . . . . . . . . . . . . . 32 Mapping of mobility system properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Experimental results of Layer 2 and Layer 3 handoff delay . . . . . . . . . . . . . . . . 134 Inter-domain handoff timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 Effect of duplicate address detection during handoff . . . . . . . . . . . . . . . . . . . . 196 Results from proactive handoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Plan for completion of my research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

iv

1 Introduction

Universal wireless connectivity to communications and information has advanced the world towardsubiquitous computing. In the space of less than thirty years cell phones have become ubiquitous andwireless data access has become common. However, this access has brought with it a variety of tech-nical problems. Radio physics and power constraints, the need to reuse spectrum, economic constraintson facility placement, and service balkanization due to competitive and political factors force us to im-plement a wireless system as cells of limited range. Furthermore, cells may use very different wirelesstechnologies or provide fundamentally different services. We then need handoff mechanisms, often atmultiple protocol layers, to allow a mobile terminal to move from cell to cell and maintain the servicecontinuity.

The simple description of mobility is that as a mobile terminal moves, it releases its binding to thecell that it is leaving and establishes a binding to the cell that it is entering while preserving the existingsession. The cellular telephony community has long implemented service and technology specific mo-bility protocols that handoff voice sessions as the user moves from cell to cell. Because voice servicequality is highly sensitive to service interruptions, cell to cell handoffs in a cellular environment havebeen highly optimized and are not noticed by the public. Tripathy et al [1] provide some of the handofftechnologies associated with cellular mobility.

For IP traffic the IETF has defined mobility protocols for both IPv4 and IPv6 [2], [3]. However, IPis dramatically more diverse than cellular voice in the range of link layer technologies used to supportIP, the number of economic units supplying IP service, the authentication protocols and services runningabove IP. This diversity meant that the IETF could not easily design into the mobility standards thehandoff optimization seen in cellular voice. As a result, standard mobile IP handoffs can take fewseconds to perform a handoff and degrades the desired quality of service in the process.

IP’s transformation from a service supporting email and file transfer to the base layer for networkconvergence means that the constraints on handover performance are becoming much more stringent.Handovers cannot interrupt real-time services. The mechanisms and design principles needed for build-ing optimized handovers in the context of mobile Internet services are poorly understood and need betteranalysis.

This thesis proposal contributes to a general theory of optimized handover, especially with respectto mobility of Internet-based applications. The contributions fall into four categories:

1. Identification of fundamental properties that are rebound during a mobility event. These propertiesprovide a systematic framework for describing mobility management and the operations that areintrinsic to handover.

2. A model of the handover process that allows performance predictions to be made for both anun-optimized handover and for specific optimization methodologies.

3. A series of technology specific optimization methodologies that could be applied to link, network,and application layers and preserve the user experience by optimizing a handover. The implemen-tation and experimental evaluation of the optimizations.

4. Use of the model to describe the optimizations and comparison of the model results against theexperimental data.

This proposal is organized as follows. Section 2 introduces the related mobility protocols, the associ-ated optimization techniques, and highlights the issues that provide the motivation for this work. Section3 provides an introduction to mobility management with an emphasis on handoff. I describe the per-formance metrics that are affected during handover in Section 4. A systematic analysis of the mobilityevent is provided in Section 5. I introduce the formal Petri net model that can analyze a mobility event inSection 6. In Section 7, I describe some key mobility optimization methodologies and demonstrate theirvalidation by way of experiments and simulation. In Section 8, I provide a roadmap towards completionof my thesis with a list of future work and timeline. Finally, Section 9 concludes my thesis proposal.1

2 Related Work

In this section, I briefly describe some of the mobility protocols and their related optimization for bothinteractive and multicast streaming traffic. I also describe some related work in the areas of modelingmobility related protocols.

2.1 Mobility protocols for interactive traffic

Current mobility management techniques can be implemented at several layers of the protocol stack,such as networking layer, transport layer and application layer. As part of my thesis, I have made asurvey of the related mobility protocols and have published the survey and the issues in [4] and [5]. Ibriefly discuss some of these protocols in this section.

Table 1 shows a qualitative comparison of some of the available mobility management protocolsin the wireless Internet. The mobility protocols with an “*” next to it are the candidate protocols thathave been developed or enhanced and have been experimented as part of this thesis. I provide a briefdescription of these mobility protocols.

Mobile IP (MIP) is a mechanism developed for the network layer to support mobility [2]. HoweverMobile IPv4 introduces network elements such as home agent and foreign agent and suffers from trian-gular routing and extra IP-IP encapsulation [6], [7] of 8 or 20 bytes. Extra overhead and longer traversalpath cause performance degradation. Mobile IPv4 usually works in two different modes, foreign agentmode and co-located mode. In co-located mode new address in the foreign network is obtained via pro-tocols like DHCP (Dynamic Host Configuration Protocol) or its faster variants such as DHCP with rapidcommit option [8].

Mobile IP with Location Registers (MIP-LR) is another network layer scheme developed to avoidencapsulation of packets [9] and to provide survivability in an ad hoc network such as military networks.It does so by replicating multiple location registers (LRs). Address management is done by DHCPservers and HLRs (Home Location Registers) provide the location updates to each corresponding hostwishing to communicate with any mobile user in the beginning. As part of this proposed work, anapplication layer MIP-LR has been prototyped.

Cellular IP [10] and HAWAII (Handoff Aware Wireless Access Internet Infrastructure) [11] are net-work layer micro-mobility management protocols. These take care of Mobile IP’s inefficiency by sup-porting intra-domain mobility and host based routing. Both of these approaches separate local and widearea mobility (i.e., adopt a domain-based approach) and use Mobile IP for inter-domain (wide area)mobility.

TeleMIP (Telecommunications-Enhanced Mobile IP) is an intra-domain mobility framework thatuses two layers of scoping within a domain and is based on IDMP (Intra Domain Mobility Protocol)[12]. By specifying an intra-domain termination point called mobility agent (MA) it helps to reduce thesignaling updates due to movement within a domain and thus reduces the loss of transient traffic dueto frequent hand-offs within a domain. As part of this proposal fast-handoff techniques for IDMP havebeen designed.

Mobile IPv6 [3] provides network layer mobility framework for IPv6. Since address autoconfigura-tion is a standard part of MIPv6, MH will always obtain a COA that is routable to the foreign network.Thus there is no need to have an FA in MIPv6 framework. When the mobile node moves to a new foreignnetwork it acquires a temporary care-of-address using stateless auto-configuration [13] or DHCPv6 [14].

There are a few transport layer mobility solutions. TCP-Migrate approach [15] proposes a new setof migrate options for TCP which provide a pure end-system alternative to network layer solutions.With this extension, established TCP connections can be suspended by a TCP peer and be reactivatedfrom another IP address without a third party (except for the involvement of dynamic DNS updates).However this approach requires modifying the transport protocol at the end-terminals. MSOCKS [16] isanother transport layer solution that introduces proxy in the middle of a network and is built on the topof SOCKS protocol [17] for firewall traversal. Upon movement of the mobile and its address change, theintermediary proxy helps splice the TCP connection. The recently developed transport protocol SCTP

2

Table 1: Survey of mobility management protocolsMobility

Type

"O" - YES, "--" - NO

IDMP **

**

**

LR

Inter-domainEncapsulation

Changes toend Systems

Trianglerouting

InfrastructureChangeProtocol

MobilityEncapsulationIntra-domain

Mobile IPV4 *

TCP-Migrate

HAWAII

MIP with

MIPv6 *

MIP-RO

MIP-RR

MIP with

FA assisted

SIP

Fast-Handoff Layer

Network

Network

Network

Network

Network

Network

Network

Network

Network

Transport

Transport

Application

Macro

Macro

Macro

Macro

Macro

Macro

Macro

Micro

Micro

Macro

Macro

LR

MSOCKS

O O O -- ---- -- O -- -- O

O O O -- -- --

O O -- O -- O

O O-- O O O

O O -- O O O

-- -- O -- -- O

-- O -- O O O

--O

O O O O

-- -- O -- O --

-- -- O -- O --

-- -- -- -- -- O

Cellular IP **

O

Macro

(Stream Control Transport Protocol) [18] has a built-in ADDIP feature that helps support continuitywhen the mobile’s IP address changes.

Application layer mobility uses the Session Initiation Protocol (SIP) [19] as the signaling mechanism[20]. This mechanism does not depend upon the home agent or foreign agent in the middle of the networknor does it require changes in the end hosts. Thus, it will help easier deployment of mobility managementsolution for wireless Internet.

MOBIKE (IKEv2 Mobility and Multihoming) [21] is an extension to IKEv2 that provides the abilityto deal with a change of an IP address of an IKEv2 end-point. HIP (the Host Identity Protocol) [22]defines a new protocol layer between network layer and transport layer to provide terminal mobility in away that is transparent to both network layer and transport layer.

2.1.1 Mobility optimization for interactive traffic

There are several enhancement to layer 3 based intra-domain mobility management solutions such asHMIP [23], [24], [25], [26] that help reduce the transient data loss when a mobile host moves betweenthe subnets within a domain. Similar fast-handoff mechanism has also been proposed for Mobile IPv6[27] that introduces an agent called MAP (Mobility Anchor Point) to localize the intra-domain mobilitymanagement. FMIPv6 provides a way of reducing the packet loss by way of fast binding update. Vakilet al [28] provide a virtual soft-handoff approach for CDMA-based wireless IP networks. It takes intoaccount the fact that both the access points receive the stream during mobile’s movement. However thisscheme does not provide a generalized solution suitable for other type of access network such as 802.11.Moore et al [29], Han et al [30] and Forte et al [31] describe some of the techniques needed to carry outDAD (Duplicate Address Detection) optimization for IPv6 clients. Park et al [8], and McAuley et al [32]describe ways of faster IP address acquisition for IPv4 clients respectively. But many of these solutionsrequire kernel modification or introduction of additional network element. As part of this work, I havedeveloped a router assisted duplicate address detection method. Details of this technique can be foundin reference[33].

2.2 Mobility protocols for multicast streaming

Mobile multicasting has primarily been divided into two categories, such as, home subscription-basedsolutions and remote subscription-based solutions.

3

Mobile-IP based bi-directional tunneling solution is actually home subscription-based solution. Itputs the multicasting burden on the Home Agent (HA) by creating tunnels between the HA and themobile using IGMP. However, tunneling multiple multicast packets to the foreign network is inefficient.Mobile Multicast (MoM) [34] proposes to reduce the explosion problem in bi-directional tunneling byelecting one designated HA. Range-based MOM [35] takes MoM approach one step further and elects amulticast agent close to FA to tunnel multicast packets to the foreign network.

Remote subscription approach takes the burden off the home agent and does eliminate tunneling byavoiding the duplication of multicast packets being tunneled to foreign networks. However this requiresthat after each handoff the user must rejoin a multicast group. In addition, the multicast trees usedto route multicast packets will be updated after every handoff to track the multicast group members.Romdhani et al [36] have discussed some of the challenges associated with multicasting to the mobileusers. It has provided a comparison of available protocols that can support mobile multicast using boththe approaches. References [37], [38] and [39] describe many of the architectural issues associated withmobile hosts in a multicast environment.

2.2.1 Mobility optimization for multicast streaming

Wu [40] proposes a solution of handover with pre-registration by deploying mobility support agents.Mobile Multicast Proxy [41] has proposed a multicast proxy approach to reduce the transient data loss ofmulticast communication during handoff. In this situation the proxy’s clients do not themselves directlyparticipate in the multicast tree. In order to limit the tree updates or limit duplication of multicast packets,proxy or agent-based solutions have been proposed in [42], [43].

2.3 Systems model for mobility

While there are several mobility protocols defined at different layers and relevant optimization tech-niques, there have been little or no work to define a formal mechanism that can represent a mobilityevent and formal optimization methodologies. However we cite few examples that have attempted tomodel certain wireless access characteristics and Mobile IPv6. Marshan et al [44] have used a Petrinet based model to analyze the performance characteristics of wireless Internet access for GSM/GPRSsystem. Amadio et al [45] have modeled the IP mobility using process calculi approach and has appliedthis into a specific protocol such as Mobile IPv6. The above work has used a software agent approach tomodel a mobility. None of these works however has actually attempted to model the intrinsic operationassociated with a handoff or has defined any formal optimization technique associated with a mobilityevent.

Following are some of the issues that are not addressed by the related work and hence create themotivation for my research.

1. Current mobility management techniques provide an ad hoc solution without any underlying sys-tematic framework and do not formalize different states and transition processes involved duringa mobility event.

2. There has not been a systematic mobility systems model that can analyze the systems optimizationtechniques for any cellular or IP-based mobility management protocol.

3. Existing mobility optimization mechanisms are tightly coupled with specific mobility manage-ment protocols. For example, it is not possible to use mobility optimization mechanisms designedfor Mobile IPv4 [2] or Mobile IPv6 [3] for MOBIKE [21]. Thus it is desirable to have a set offormal mobility optimization methodologies with specific design rules or methodology that canwork with any mobility management protocol and access technologies.

4. Existing works do not provide a generalized mobility optimization framework that can supporthandoff involving both single interface and multi-interface terminals.

4

5. There is no existing mobility optimization mechanism that easily supports handovers across ad-ministrative domains without assuming a pre-established security association between administra-tive domains.

This thesis proposal addresses some of these issues. It provides a detailed analysis of the systemsproperties associated with a mobility event, develops a formal model based on these properties, and de-fines several systems optimization techniques. Then it validates these techniques by way of simulation,experiments and numerical analysis using a few network layer and application layer candidate mobilityprotocols. A detailed analysis of these properties help to formulate an abstract mobility systems frame-work that consists of different components associated with a handoff process. I formalize the mobilitymodel by representing these processes in terms of communicating Finite State Machines (FSM) andmodel the associated states and transitions using Timed Petri net. Based on the analysis of these sys-tem properties I propose several reactive and proactive systems optimization techniques that can providedesired performance metrics for both interactive traffic such as VoIP and multicast streaming traffic. Iapply some of these techniques to few candidate network layer and application layer mobility manage-ment protocols and compare the results for both interactive and streaming traffic. I also apply theseoptimization methodologies to the Petri net model for verification.

3 Mobility Management Introduction

In this section, I introduce mobility management with a focus on handoff management. A mobilitymanagement scheme primarily consists of two components: location management and handoff man-agement.

Location management enables the network to discover the current point of attachment of the mobileuser so that the new connection can be established when a new call arrives.

Since my work is focused on handoff management, I elaborate the handoff management in this sec-tion. Handoff management allows the network to maintain the user’s connection binding as the mobilemoves from one attachment point to another in the network. As a mobile goes through a handover pro-cess, it is subjected to handover delay because of the rebinding of properties at several layers of the proto-col stack. There are several common properties that contribute to the re-establishment of the binding pro-cess or modification of these layers. These properties can mostly be attributed to factors such as accesscharacteristics (e.g., bandwidth, channel characteristics), access mechanism (e.g. CDMA, CSMA/CA,TDMA), re-configuration of parameters at different layers, re-authentication, re-authorization, rebindingof security association at all layers. During any handover procedure any multimedia application runningwithin a client gets affected because of the delays incurred within each of the layers of the protocol stack.

A handoff process can be either network controlled handoff, mobile-assisted handoff or mobile-controlled handoff. Handoff management primarily involves five main phases, namely, 1) Decision 2)Discovery 3) Selection 4) Configuration and 5) Execution. Each of these phases includes several sub-phases within it. In the first phase, either the mobile or the network agent determines the need for handoffbased on the network condition, signal-to-noise measurement or certain metrics based on policy that canbe either local or in the network. The second stage is the discovery phase where the mobile or networkdiscovers possible new resources for the handoff connection. This phase involves discovering both theneighboring networks and the resources within the network. Once the probable networks and resourcesare discovered, the mobile selects the new network either by itself or with some help from the network.Fourth phase is the configuration phase. This phase actually helps preparing the re-connection path ofthe mobile. During this phase the mobile connects to the new point of attachment and establishes its newconnection identifier by performing any re-routing operation. Any desired authentication or encryptionoperation is also included in this phase. The final phase is the execution phase. This phase involvesdata flow control where the delivery of data from the old connection path to the new connection path ismaintained according to the pre-defined service guarantee.

The handover process can be primarily classified into homogeneous and heterogeneous handover.5

Multiple Interface Case (802.11b – CDMA1XRTT) – MIP as mobility

802.11 802.11CDMAHandoff

19 s

Single Interface Case (802.11b – 802.11b) – SIP as mobility

802.11 802.11Handoff4 s

Handoff17 s802.11 CDMA 802.11

Multiple Interface Case (802.11b – CDMA1XRTT) – SIP as mobility

Figure 1: Effect of handoff delay due to mobility event

Heterogeneous handover includes movement between various types of wireless networks and differentadministrative domains. Supporting terminal handovers across heterogeneous access networks such asCDMA (Code Division Multiple Access), IEEE 802.11, WiMAX (IEEE 802.16) and GPRS (GeneralPacket Radio Service) needs to take into account different QoS, security and bandwidth characteris-tics associated with each type of access network. Similarly, movement between two different kindsof domains poses a challenge since a mobile will need to re-establish authentication and authorizationas well in the new domain where each administrative domain may or may not have any prior securityarrangement.

An efficient flow control mechanism reduces the information loss during the handover by introducingdifferent operations such as buffering, copy-forward and multicasting techniques. Some of the researchissues with handoff management include efficient and expedient packet processing, minimizing signalingload, optimizing the routing for each (re)connection, optimized re-configuation, efficient bandwidth re-assignment, minimizing the packet loss and delay during handover. However there has not been anyformal analysis nor system model developed to study the system optimization techniques associatedwith the handoff during a mobility event.

4 Handoff Performance Metrics

Systems optimization mechanisms associated with any mobility management scheme need to considerseveral performance metrics that affect the perceived quality of service for any real-time application. Inthis section, I introduce some of these performance parameters that get affected due to mobility. Figure1 shows how a real-time communication is affected while the mobile is subjected to heterogeneous andhomogeneous handoff under a normal mobility scheme without any optimization technique. There areseveral intrinsic operation that contribute to this delay and associated packet loss. Thus it is desirableto make a formal analysis of the intrinsic steps that constitute the handoff process and develop theassociated optimization methodologies.

End-to-end delay and packet loss tolerance level may vary from application to application. In orderto provide the desirable quality of service for interactive VoIP and streaming traffic, one needs to limitthe value of end-to-end delay, jitter and packet loss to a certain threshold level. Based on the type ofapplication (e.g.,interactive, streaming, data) different standards organizations define different thresholdlimit for these metrics. For example, for one-way delay, ITU-T G.114 recommends 150 ms as the upperlimit for most of the applications, and 400 ms as generally unacceptable delay. One way delay tolerancefor video conferencing is in the range of 200 ms to 300 ms. Also if an out-of-order packet is receivedafter a certain threshold it is considered lost. Similarly, a normal voice conversation can tolerate up to 2

6

percent packet loss.

1. Delay and jitter

A mobility event contributes to two kinds of delays that are of interest such as, end-to-delay of apacket and delay between the last packet in the old point of attachment and first packet in the newpoint of attachment.Typically end-to-end delay during the traversal of a packet consists of several components, suchas network delay, operating system (OS) delay, CODEC delay and application delay. Networkdelay ( ������� ) consists of transmission delay, propagation delay, queueing delay in the intermediaterouters. Operating system related delay consists of scheduling behavior of the operating systemin the sender and receiver. CODEC delay ( ������ ) is generally caused due to packetization anddepacketization at the sender and receiver end. Application delay is mainly attributed to playoutdelay that helps compensate the delay variation within a network. This is often referred to asdejitter delay ( ������� ). In case of interactive VoIP traffic, Mouth-To-Ear delay and distortion arethe important factors that affect the overall quality. Based on the above components, a typicalMouth-To-Ear (M2E) delay for VoIP traffic can be represented as follows����� = ������ + ������� + ������� + ��� �� + ��!��#" + �$�����

On the other hand, handover delay refers to the time elapsed between the last packet received fromthe old point of attachment and first packet received in the new point of attachment. Ideally, latencyintroduced due to handoff gets added to the variable part of the network delay � �%�&� . Typicallyhandoff delay resulting out of change in the network path is contributed due to the rebindingdelay at different layers of the protocol stack, binding update delay, delay due to authentication,security association, media redirection delay. Binding update delay may depend upon networktransmission delay, processing delay at the end points, number of message exchange between themobile and network node. A comprehensive analysis of delay components arising out of handoffis given in Section 6.Jitter refers to the variation in consecutive packet arrival time at the receiver side caused due tovariation of end-to-end delay of any packet during handoff. Buffering and queueing delay in themiddle of the network also adds to jitter value.

2. Packet loss

Packet loss in a communication path is typically caused by congestion, routing instability, linkfailure, lossy links such as wireless connection. However during a mobile’s handover, a mobileis subjected to packet loss because of its change in the attachment to the network. Lost packetscontribute to congestion in case of TCP traffic because of re-transmission. On the other handfor both streaming traffic and VoIP interactive traffic packet loss will affect the service quality.Number of packets lost depends upon the handover delay and rate of traffic at the sender side.While reducing the handover delay helps reduce the packet loss it is also essential to mitigatethe effect of handover delay by reducing the packet loss during handover delay. Buffering at thenetwork edge, FEC (Forward Error Correction), copy-forward mechanism and bicasting are someof the techniques that help mitigate the effect of handoff delay by reducing the packet loss at thecost of additional delay.

3. Location update overhead

Location update overhead refers to the number of control messages for location management andhandover functions. By limiting the number of messages and confining its traversal boundary, ithelps to optimize the delay due to location update. An optimized location update helps to reducethe number of missed calls during handoff.

4. Throughput7

L3 Attachment PointRAP

R

Domain Router

DomainRouter

CH

R

DomainRouter

DomainRouter

MHAP

AP

AP

CORE Router

L2 Attachment Point

CORE Router

AP

AP

AP

AP

AP

AP

R

R

RR

AP

AP

R

R

RR

R

R

R

R

R

R

AP

AP

AP

AP

AP

AP

R

R

R

R

R

AP

AP

AP

AP

AP

R

Figure 2: Infrastructure-based mobility

Throughput is a measure of the amount of data received by the mobile over a certain period oftime. While network access characteristics such as bandwidth affects the throughput, handoffoperation also contributes adversely to the overall throughput of the end-to-end communicationbecause a new data path is established as a result of handoff leads to data interruption. Overallthroughput is dependent upon packet loss and packet arrival delay in the system.

5 Systems Analysis of Mobility Event

In this section, I analyze different components that are part of a mobility event and highlight the commonproperties that are associated with this event. Analysis of these properties helps provide a systematicapproach to the design of a formal mobility model that is described in Section 6.

Figure 2 shows the network nodes associated with infrastructure-based mobility. It shows all thenetwork components that operate at different layers for routing packets from source to destination. Aspart of this analysis, I investigated and analyzed several cellular and IP-based mobility protocols. Inparticular, I study the handoff process associated with the cellular protocols such as GSM, CDMA, LANbased 802.11, network layer mobility protocols such as Mobile IP over IPv4 and IPv6 and applicationlayer mobility protocol such as SIP. After a careful study of several of these mobility techniques, Iextrapolate certain basic functions or systems properties that occur during a mobility event. Optimizationof these functions can enhance the overall systems performance of any mobility event.

By analyzing different intrinsic operation involved in each of these schemes, I extrapolate the abstractproperties associated with a mobility event and then design a common mobility management frameworkthat consists of a set of communicating processes associated with a mobility event. Any mobility protocoland its properties can be easily represented by this common framework. For example, total number ofround-trip signaling messages during location management, packet loss, delay, jitter, amount of signalingupdate during movement are some of the parameters that can be represented by this model. The modelcan analyze these parameters and evaluate the effectiveness of any mobility protocol. I use this mobilityframework to validate the properties by way of experimental and analytical results and compare thesewith the non-optimized version. I describe below results of my analysis about the functional componentsassociated with a mobility event.

1. Decision

Based on the measurement such as Signal-to-Noise Ratio (SNR), the mobile makes a decisionabout the impending handoff and starts the neighborhood discovery process to find the best avail-able network where it can handoff to.

2. Discovery

8

Discovery phase is the second phase of any mobility event. A discovery process can consist of sev-eral sub-processes such as neighborhood network discovery, resource discovery in the neighboringnetworks where a mobile may move into as part of the handoff process.

(a) Network discovery

During the process of a handoff, the communicating mobile needs to discover the neighbor-ing networks and the associated network elements around the current point of attachment.Based on the type of the networks it plans to connect to, (e.g., 802.11, CDMA, GSM) dis-covery of the appropriate network takes certain amount of time. For example, in an 802.11b-based network the mobile performs scanning to determine the neighboring networks. How-ever during this scanning process mobile’s communication is interrupted briefly. Thus opti-mizing the network discovery process can contribute to the optimization of overall mobilityevent.

(b) Resource discovery

Once the target network is discovered, several resource parameters within the target networkneed to be discovered as well. These include channel number, frequency, bandwidth, encryp-tion algorithm, authentication server, configuration server etc. Resource discovery processhelps to configure the mobile with proper channel number, frequency and authenticationparameters in the new network.

3. Network selection

After a handover decision is made either by the mobile or by the network, the mobile needs toselect the new network and connect to the new point of attachment. Thus an appropriate selectionmechanism and faster detection mechanism to the new point of attachment will help the overalloptimization process. Network selection phase consists of the following sub-phases.

(a) Detection of new point of attachment

Detection of new point of attachment usually follows the discovery process. Based on thetype of networks, it could involve the following operation such as detection of new Loca-tion Area in case of GSM, detection of new Routing Area in case of GPRS, or detection ofsubnet, domain, cell in case of 802.11-based handoff in an IP-centric network. Each mo-bility management scheme provides different means of achieving this, such as, GSM usesBCCH, CDMA uses pilot channel and 802.11 uses beacon interval to detect new point-of-attachment. In some cases some of the upper layer detection mechanisms such as ForeignAgent advertisement, ICMP router advertisement, can help provide faster detection involv-ing layer 3 movement. Currently the IETF is working on defining a protocol for Detectionof Network Attachment (DNA) [46].

(b) Detection of loss of old point of attachment

In case of sudden lapse of coverage or disconnection it is important to detect the absence ofold connection quickly so that it can start the discovery process of the new point of attach-ment and prepare for the new connection.

(c) Event notification

An event notification such as the availability of the new point of attachment or lapse ofan existing connection is usually provided to the upper layers so that any further handoffrelated upper layer functions can take place to expedite the mobility event. These upperlayer functions include configuration, registration, binding update and media redirection.Signal-to-Noise Ratio (SNR) threshold is one possible event notification technique.

(d) Handoff triggers

Handoff trigger is the last of the events during the network selection process. Handoff de-cision to switch access networks could be mobile controlled, network controlled or mobile9

assisted. Since layer 2 association takes place before any upper layer association, it helps toprovide a layer 2 trigger before the upper layer mobility management functions get executed.Thus an optimized triggering mechanism helps to expedite the handoff process.

4. Configuration

The configuration phase associated with the mobile during its handoff involves the following sub-phases.

(a) Identifier configurationThis process allows a mobile to configure new temporary connection identifier either at layer2 or at layer 3 in the new point of attachment of the network, such as Care-of-Address incase of IP environment, TMSI (Temporary Mobile Subscriber Identity) in case of GSM.Configuration process of a new identifier requires a series of signaling handshake betweenthe mobile and the server in the network. It also involves a set of processes such as securitybindings, uniqueness testing of the new identifier that add to the delay as well.

(b) RegistrationRegistration is a process of establishing the mapping between permanent identifier and tem-porary identifier for proper location management functionality. An optimized or hierarchicalregistration process helps expedited location management and faster delivery of the newcalls.

(c) Security bindingBefore a new communication path is established between the end-points, communicatingmobile node needs to authenticate itself and then establish some security association with thenetwork elements those are in the communication path. In many of the mobility managementprotocols execution of these processes are mandatory and security association may take placeat all the layers. Establishing a security association involves exchange of signaling betweenthe mobile and any centralized security server that dispenses the key to the mobile.

(d) AuthenticationAuthentication process provides a means for the mobile to gain access to the network re-sources before the communication could start. This process involves handshake between themobile and the authentication server in the network and can happen at different layer. In caseof GSM it uses SRES and A3 algorithm for the authentication. In case of mobility in 802.11environment it uses 802.11i in layer 2 and other mechanism such as PANA (Protocol forcarrying Authentication to Network Access) at layer 3. For example, Georgiades [47] showsthat it takes up to 5.3 seconds to complete an authentication procedure that involves a com-bination of EAP (Extensible Authentication Protocol) [48], TLS (Transport Layer Security)[49] and IEEE 802.1x based authentication.

(e) EncryptionAfter a mobile is authenticated to the network, data and signaling protection are providedby means of encryption. In order to provide an encryption, the shared key needs to beexchanged between the communicating nodes or between a communicating node and thecentralized server. For IP-based mobility, encryption can take place at different layer suchas WEP at layer 2, IPSEC at layer 3 and SRTP Secured RTP at layer 4; similarly GSMuses SRES algorithm to derive the encryption key where as IS-41 uses CAVE algorithm toderive the encryption key. An encryption process adds to the handoff delay because of theassociated encapsulation, decapsulation and processing at the end host and key exchangeneeded to setup the encryption.

5. Execution

An execution phase is considered to be the last phase during the handover process. It consists ofthe following sub-phases. 10

(a) Binding updateBinding update is the process by which a mobile can associate new network identifier forrerouting the data to the new destination. As the mobile connects to a new point of attach-ment and obtains a new temporary network identifier (e.g., TMSI in GSM, COA in MIPv6,FA-COA in MIPv4) in the new network, it needs to update the communicating host or someother node such as home agent that has been part of the communicating path before thehandoff. This signaling handshake helps to reassociate the new network identifier with thepermanent identifier of the mobile but takes certain amount of time as well. Until the map-ping is complete, the transient data still goes to the old network and is considered lost inthe absence of any optimization mechanism such as buffering or packet forwarding. Thusoptimization during binding update phase can help reduce the handoff delay and associatedtransient data loss. In some cases, this binding update needs to be authenticated and securedas well. MIPv6 provides return routability procedure and adds two more messages such asCTI (Care-of-test Init) and HTI (Home test Init) to obtain the binding key so that bindingupdate can be authenticated as well. This is an additional delay. Unlike unicast case, in caseof multicast traffic which is receiver driven, a trigger mechanism to join a multicast tree isconsidered equivalent to a binding update.

(b) Media reroutingOnce the binding update is complete, data from the correspondent node gets routed to thenew node. Media redirection process involves several functions such as encapsulation, de-capsulation, buffering and tunneling at different parts of the network. These processes takecertain amount of time. Thus media re-direction adds to the delay factor as well. During themedia re-routing process transient data may get lost. Thus there is a need to optimize thisfunction or mitigate its affect. Following are some of the intrinsic operation that may affectthe efficiency of media re-routing.

i. Encapsulation

ii. Decapsulation

iii. Tunneling

iv. Buffering

v. Copy and forwarding

In Table 2, I list some of these above functions and how these are carried out for different type ofmobility protocols.

6 Systems Model for Mobility Event

In this section, I develop the mobility systems model by analyzing the state transition associated witheach layer and represent each transition using a set of finite state machines. I use Deterministic TimePetri net [50] to model the mobility event and derive the relevant optimized models by applying theappropriate rules of reduction [51] associated with Timed Petri net.

6.1 Layered approach to mobility optimization

I begin with investigating each layer and then analyze rebinding of some of these common propertiesat each layer during the handover procedure. Figure 3 shows the latency factors associated at differentlayers during a handoff contributed by certain intrinsic operation associated at each layer. Optimizationtechniques can be applied to each of these intrinsic operation at each layer to reduce the overall handoffdelay during the handover.

11

B2BUACHRTPtrans

Re-INVITE

IPSEC/SRTP/S/MIME

INVITE exchange/AAA

CoAAORRe-Register

Router Prefix, ICMP

L3RouterAdv.

802.2111R

StatelessICMP Router

AnySIPM

CHMAPHA

MIP updateMIP RO

IPSECIKE/PANAAAA

CoARouterPrefix

RouterAdv.

CARD802.2111R

StatelessProactive

AnyMIPv6

FARFAHA

MIPRegistration

IPSecIKE/PANAAAA

FA-CoACo-CoA

FA advFA adv.L2 triggering

ICMPRouterAdv.

ICMPRouter adv.FA adv.

AnyMIPv4

IntermediateyRouter

RouteUpdate

IPSecIPSecMAC AddressAP address

GW Beacon

APbeaconID

Mobilemsmt.

Gatewaybeacon

AnyCell IP

IAPPAssociateWEP/WPA802.11i

Layer 2 authenticate802.1XEAP

SSID,Channel number

Scanning.ChannelNumber,SSID

SNR atMobile

11R802.21

Beacon11R

CSMA/CA

802.11

PDSN/MSCMSCAESDiffie-Hellman/CAVE

TMSIRTCChannelStrength

SYNCChannel

PILOTChannel

EVDOCDMA1X-EVDO

AnchorMSC

MSCContld.

KasumiDiffie-HellmanAKA

TMSIRTCChannelStrength

SYNCchannel

PILOTCDMAIS-95

AnchorNetworkControl

AESSRES/A3TMSIFrequency

ChannelStrength

SYNCChannel

PILOTCDMAWCDMA

AnchorMSCContld.

DESSRES/A3TMSISCHChannelStrength

FCCHBCCHTDMAGSM

MediaRerouting

BindingUpdate

EncryptionKey exchange/Authentication

ConfigurationDetectionTechnique

TriggeringTechnique

Resource Discovery

Network Discovery

AccessType

Mobility/

Function

Table 2: Mapping of mobility system properties

Layer 2 delay: In a typical 802.11 environment, channel scanning, probing, authentication and asso-ciation are the basic set of functions that contribute to the standard delay before a mobile completes theassociation at layer 2. Encryption and user authentication using WPA (Wi-Fi Protected Access) in con-junction with 802.1X and EAP (Extensible Authentication Protocol) contributes to the additional delaybecause of 4-way handshaking. Figure 4 shows interaction of different state transition within layer 2 thatdoes not involve WPA. I have taken measurements for layer 2 handoff for different operating system andlayer 2 driver and have presented these in Table 3.

Layer 3 delay: In an IP-based environment, a layer 3 transition process goes through several steps suchas new IP address acquisition, duplicate address detection, ARP update, and local authentication. Eachof these operations involves number of message exchange between the mobile and some other entity inthe network. In some cases an authorization process is also involved even before a new IP address isobtained. There are several protocols such as DHCP, DHCPv6, PPP or stateless autoconfiguration thatare used to take care of IP address acquisition. In a typical inter-domain mobility scenario independentof inter-technology or intra-technology handoff, introduction of AAA functionalities adds an additionaldelay component, as the mobile requests access to network services. These AAA transactions includea series of steps such as re-authentication, re-authorization between the mobile and the AAA serversduring the handoff. Depending upon the type of architecture, in some cases the AAA signals traverseall the way to the home AAA server before the network service is granted to the mobile in the newnetwork. Thus it is desirable to have an optimized method of interaction between the MH and AAAservers during the handoff. Each of these above steps may take different amount of time for a layer 3transition to complete and thus can benefit from optimization.

Application layer delay: After a layer 3 transition is complete and layer 3 identifier is reconfigured,there are upper layer functions such as the binding update from the mobile, media redirection at thecorrespondent host, establishment of upper layer encryption such as TLS, SRTP that need to be com-pleted. Delays due to these functions can be attributed to transport delay, processing delay at the endhosts. In case of multicast traffic, the triggering mechanism at the receiver that helps the mobile to jointhe multicast tree is the source of delay.

12

Intrinsic Operation

L2 Delay

L 2 Scanning

Association

L2 security

L3 Delay

Address Acquisition

DuplicateAddress Detection

ARPUpdate

Local Authentication

AAAProfile

BindingUpdate

Media RedirectionApplication

LayerDelay

Figure 3: Handoff latency components

Table 3: Experimental results of Layer 2 and Layer 3 handoff delay

Based on the type of mobility, different layers are affected. For example, in the inter-subnet mobil-ity category which is intra-domain there is no additional delay due to authentication and authorizationprocess that is usually required during inter-domain mobility.

I have taken several measurements to study the delay associated with layer 2 handoff and layer 3IP address acquisition. Table 3 provides a snapshot of some of these experimental results associatedwith layer 2 and layer 3 handoff operation under different scenario. These values may vary dependingupon the operating system (e.g., Linux, Windows) of the end clients and the associated drivers. Theexperimental results associated with L2 handoff delay does not include WPA or EAP authentication. Amore detailed analysis of layer 2 handoff analysis shows that scanning and probing operation contributeto the most of the delay. For example, using active scanning with Orinoco driver in a Linux environment,it takes almost 78 ms for probing action to complete followed by layer 2 authentication and associationthat take about 2 ms and 20 ms respectively.

6.2 Petri net model for mobility event

The mobility event is viewed as the perturbation to the steady state of a communicating node that mayaffect different layers in the protocol stack. As a communicating node is subjected to a mobility event,it goes through certain set of intermediary states before it attains steady state again by returning tothe connected state. In some cases because of the ping-pong effect the mobile can oscillate betweenconnected and disconnected states. As described in the Section 5, several common processes such astriggering event, resource discovery, reassociation of network identifier, configuration, authentication,secured rebinding update, media redirection that occur during this state transition. Depending upon thetype of mobility each layer in the protocol stack gets affected and the mobile goes through a series of

13

State1 UnauthenticatedUnassociated

State 2AuthenticatedUnassociated

State 3AuthenticatedAssociated

SuccessfulAuthentication

SuccessfulAuthentication orRe-association

DisassociationNotification

De-authenticationNotification

De-authenticationNotification

Class 1Frames

Class 1 & 2Frames

Class 1, 2 &3Frames

Class 1 Frames – Control FramesClass 2 Frames – Management FramesClass 3 Frames – Data Frames

Figure 4: Finite state machine model L2 transition

similar transition processes within each layer. As a result of series of state transitions involved duringthe mobility event, a mobile’s communication is interrupted because of the delay associated with eachtransition. Any mobility optimization mechanism will help speed up this transition process from onestate to another. Since most of these state transitions during the mobility event appear to be sequentialin nature, the mobility system exhibits similar behavior often observed by the Flexible ManufacturingSystem. There have been works by Zuberek et al [52] to model and analyze the simple schedules formanufacturing cells. Performance analysis for the mobility systems model can apply similar techniquesas well to find out efficiency of the reduced systems model.

connected

Disconnected

Discovered Selected Configured

AuthenticatedEncryptedUpdated

NetworkResourceDiscovery

NetworkSelectionDetection

NetworkConfiguration

MobileAuthentication

EncryptionBinding Update

Forwarding

P1 P2 P3

P4P5P6

P0

t1 t2 t3

t4

t5t6t7

P7

Intra-domainBinding update

t8t9

BufferingRedirection

t0

Figure 5: A generalized Petri net model for mobility event

Thus overall mobility event can be thought of as a series of sequential states, and there are severalsub-events within each of these states. When the mobile is subjected to a series of consecutive handoffs,it can be modeled as a cyclic structure. Figure 5 shows how the state machines associated with a mobilityevent can be put into a formal framework by using Timed Petri net approach [50]. Each place (P)represents various stages of the mobility event and the transition (t) represents the time taken due todifferent set of operations between the stages. Each of these stages can be considered as sub-system.The Petri net model representing the general mobility systems is actually a decision free Petri net, wherea minimum cycle time is an indicator of maximum performance. The cycle time is represented as

C = max �� / � :k=1,2,3...q, where �# = sum of the execution times of the transmissions in circuit kand ( � ) is the total number of tokens in the places in circuit k and q is the number of circuits in the net.

In case of a systems model involving mobility event, these values can vary depending upon the num-ber of transitions involved in a cycle. The optimization methodologies can be applied to the appropriatestates of the general mobility model to help derive the reduced version of the corresponding model. For

14

example, Figure 5 also introduces a new place P7 and thus an alternate transition path, that providesa faster transition to the connected state by allowing local buffering and media redirection. In this ex-ample, I have taken advantage of the lateral fusion reduction rule of Petri net to represent this specificoptimization method.

connected

Disconnected

Layer 2 Layer 3 ApplicationLayer

Scanningselection

L2Authenticationencryption

L2Association

MeasureNetworkStrength

Set upARP

L2Trigger

P1

P2

P3

AddressAcquisition

AddressResolution

L3AuthenticationEncryption

BindingUpdate

t00

t12

t1

t2

t3

t23 t4

t5 t6

t7L3Layer trigger

AppLayerencryption

forwarding

t34

t8

P4

P5 P6 P7

P8 P9 P10

P0

Figure 6: Petri net model with layered events

Figure 6 shows Petri net representation of several state transitions at each layer. It illustrates thecomponent level interaction in each layer. Several subprocesses in each of the layers such as scanning,layer 2 authentication, layer 3 address acquisition, layer 3 authentication, encryption, binding update areillustrated here.

It is evident that a mobile is subjected to different amount of handoff delay based on the types ofhandover it is subjected to. For example an intra-subnet, intra-technology handover will take much lesstime than inter-subnet and inter-technology handover, because the mobile goes through far less numberstate transitions during the handoff process. Reduction of time during state transition, parallelizationof some of the state transitions within each layer, reduction of number of states by way of fusion willcontribute to the overall optimization to the handoff.

7 Optimization Methodologies and Experimental Validation

In this section, I elaborate some of the key optimization methodologies that could be applied to a mo-bility event and validate these techniques by means of experiments and simulation. These can primarilybe categorized into Reactive, Proactive, Multilayer. I apply these techniques to both interactive andmulticast streaming traffic and extend it to simultaneous mobility as well.

Figure 7 illustrates the Internet multimedia testbed that I have implemented to conduct our exper-imental validation. It shows how a mobile user keeps moving from its home base in the Internet, andis subjected to different kinds of hand-off such as micro (cell), macro (subnet), and domain handoff asit moves between heterogeneous networks. Each domain is equipped with network elements such asmobility servers, QoS servers, AAA (Authentication, Authorization, Accounting) servers, SIP servers,and DHCP servers that provide the desired roaming functionalities. I describe the details of the imple-mentation for the wireless Internet architecture in [53].

Each of these methodologies described below can be considered as set of fundamental principles ofoptimization for a mobility event.

7.1 Reactive methodologies

In this section, I introduce few key rules of mobility optimization that fall under reactive category anddescribe the associated experimental results that validate these techniques.

15

7.1.1 Maintain direct media path between CH and MH

It is desirable to avoid the triangular routing and encapsulation mechanism associated with any mobilityprotocol as it contributes to the performance degradation by adding delay and data overhead. In this sec-tion, I describe the optimization mechanism that helps to maintain the direct path between correspondenthost and mobile host thereby avoiding triangular routing and any encapsulation overhead. I validate thisoptimization technique by implementation and performance analysis of the following candidate mobilityprotocols and comparing these with MIPv4.

1. SIP-based subnet mobility

2. SIP-based inter-domain secured mobility

3. Application layer MIP-LR

Figure 7: Experimental mobile multimedia testbed

Originally SIP-based mobility was proposed by Schulzrinne and Wedlund [54]. I augment the SIP-based terminal mobility with a complete hand-off process [5] as part of this thesis. The augmentationincludes a combination of network detection, registration, configuration, dynamic address binding, se-curity, location management functions, and roaming support for wireless Internet [5]. In the followingparagraphs I present the simulation and experimental results from application layer subnet and inter-domain mobility, application layer MIP-LR and compare these with network layer mobility such asMIPv4.

SIP-based subnet mobility: Figure 8 shows latency comparison between SIP-based and MIP-basedterminal mobility during a subnet handoff. Curves show the relative performance improvement of SIPover Mobile IP for different packet size analyzed both from NS2 based simulation and laboratory ex-periment. SIP-based terminal mobility reduces the delay by avoiding the triangular routing inherentlypresent in basic Mobile IPv4. By using SIP for mobility management one can expect to have 50 per-cent latency improvement in real-time (RTP/UDP) traffic thus providing a reduction in latency from abaseline of 27 ms to 16 ms for large packets and a 35 percent utilization increase 60 bytes/packet sizecompared with baseline of 80 bytes/packet size with IP-in-IP encapsulation in Mobile IP. These resultsdemonstrate that how maintaining direct media path between CH and MH helps provide the optimizationin terms of lower end-to-end delay and less overhead.

16

Figure 8: End-to-end delay improvement using direct path

Inter-domain secured terminal mobility: Inter-domain mobility introduces new operation such asestablishment of authentication with the local access servers and interaction with an authorization serverafter the mobile moves to the new domain. As a result, this process will introduce additional delay to thehandoff process. In order to realize a secured inter-domain mobility and study the effect of security andauthentication on the handoff process, I have designed and implemented a multi-layer security frame-work based on SIP-centric architecture. It provides access control to the network using PANA [55],profile verification using Diameter [56], last-hop-over the air protection using packet based encryptionsuch as IPSec. End-to-end encryption for multi-media traffic (e.g., audio, video) is provided by SRTP(Secured RTP) [57] an application layer encryption. SRTP key is distributed securely using INVITEexchange and S/MIME [58]. I experimented inter-domain mobility by using both network layer (e.g.,Mobile IP) and application layer (e.g. SIP-based) mobility techniques and have compared the experi-mental results. I provide the details of these experiments in [53] and [59]. But I provide some sampleresults here.

Domain handoff delay is comprised of several components such as delay due to 802.11b channelchange, subnet and domain discovery, IP address acquisition, local authentication by means of PANA,profile verification using AAA diameter, and delay due to SIP re-INVITE or MIP registration. In SIP-based mobility experiment, a complete Re-INVITE, OK and ACK sequence took about 500 ms includingthe processing time at the end hosts. In the specific laboratory environment it took about 100 ms pro-cessing time between the messages at the end host, and 70 ms for each message to traverse between theend points. Address acquisition time due to message traversal of DRCP messages in a specific subnet isapproximately 100 ms. However it does not reflect the extra time needed to check the layer 2 channelchange (typically 100 ms for 802.11b access point ) or interval of periodic DRCP server advertisement.A typical SIP registration to update the client’s new IP address in the SIP registrar takes about 150 ms.It is noteworthy to mention that the delay parameters strongly depend on the media used, number ofrouters in the path, background traffic, number of hops, authentication methods and processing speedof the correspondent and mobile hosts. Table 4 shows the timing associated with each of the functionalcomponents during inter-domain handoff using both SIP and MIP as the mobility protocols. Some ofthe end systems processing time are not shown here. RTP1 represents the time when the RTP traffic isreceived prior to handoff in the old network and RTP2 represents the time RTP traffic is received in thenew point of attachment after the handoff. For both the cases IKE (Internet Key Exchange)took almostclose to 5 seconds before the security association is established.

Application layer MIP-LR: I have designed application layer MIP-LR that augments the basic MIP-LR scheme with some application layer techniques such as libipq, mangler and NAT modules and thus

17

Table 4: Inter-domain handoff timing

Mobility RTP1 Layer2 Router adv Layer3 Auth IKE BU RTP2 DelaySIP 51.756 s 120 ms 500 ms 80 ms 10 ms 5130 ms 240 ms 53.066 s 1.31 sMIP 23.806 s 120 ms 500 ms 80 ms 10 ms 4580 ms 20 ms 31.186 s 7.380 s

provides kernel level independence. Like SIP-based mobility protocol, MIP-LR also tries to maintainthe direct path between CH and MH during the handoff by providing direct binding update technique.From the experimental results in a laboratory prototype I have verified that on can attain up to 50 percentreduction in management overhead and up to 40 percent improvement on latency compared to standardmobile IP in co-located mode. Figure 9 shows round trip time delay comparison between applicationlayer MIP-LR and Mobile IP. This result also verifies that direct path between CH and MH providesoptimization.

MIP-LR vs. MIP Delay (Experiment)

4

6

8

10

12

14

16

18

20

0 100 200 300 400 500 600 700 800 900 1000 1100

Bytes per packet

Ro

un

d t

rip

tim

e in

Mse

c

MIP

MIP-LR

Figure 9: Round trip delay comparison - MIP-LR vs. Mobile IP

7.1.2 Minimize layer 3 configuration process

In this section, I demonstrate how an expedited layer 3 configuration can provide mobility optimizationand discuss the factors that may affect this process. I have investigated the effect of layer 3 configurationon the handoff for both IPv4 and IPv6 networks. IPv6 provides a different layer 3 configuration anddetection mechanism compared to IP4. I describe some of the experimental results using both statefulmode in IPv4 and stateless mode in IPv6 and analyze several components such as duplicate addressdetection and router selection that contribute to the delay. Then I apply optimization techniques andvalidate these with the experimental results.

I have experimented with real-time traffic over IPv6 network using both Mobile IPv6 and SIP-based terminal mobility. I have used Linux kernel version 2.4.9 with patch from USAGI projects(www.usagi.org) in the routers and Linux hosts. I adopted MIPL Mobile IPv6 for Linux [60] to sup-port mobility in the testbed. I have carried out several experiments to analyze the effect of DuplicateAddress Detection (DAD) [13] in the disruption of real-time voice traffic. Results from an experimen-tal handoff analysis of SIP mobility with IPv6 and MIPv6 involving signaling and media redirection isshown in Table 5. Timing for SIP mobility was measured for both DAD and No-DAD cases where asonly no-DAD case is shown for mobile IPv6. I added aggressive router selection procedure as part of thekernel module for MIPL that forces a mobile to bind to the new router quickly enough without doing aNeighbor Unreachability Detection [61]. We have shown the details of the experiment in [62]. From theexperimental results, I infer that by getting rid of DAD and adopting aggressive router selection processwe were able to minimize the signaling delays to 200 ms and media delays to less than 500 ms.

18

Table 5: Effect of duplicate address detection during handoff

CASE

H23

SIP with DAD

H12

SIP with DAD

HANDOFF Media

SIP w/o DADSIP w/o DAD

(ms) {ms)

1934.7 161.1

4187.7

(Home - Visited)

(Visited - Visited)

(Visited - Home)

161.6

171.4

1.0 H31

2.0

1.5

1949.4 408.4

418.6

420.8 21.1

30.3

25.3

3932

MIPv6 w/o DAD MIPv6 w/o DAD

Signaling

3829 3854

Similarly, as part of my experiments with DHCPv4 and MIP, I found that, with ARP check enabled,the IP address acquisition took an average of 15.18 seconds, but when the ARP check is suppressedaverage time taken for IP address acquisition was 436.75 ms. There are few proposals in the IETF suchas Passive DAD and DHCP rapid commit option that try to expedite the IP address acquisition. As partof the initial work I have also implemented a router assisted duplicate IP address detection mechanismmeant for DHCPv4 environment that reduces the layer 3 configuration. In this case, router or a specificserver provides the list of the used IP addresses in the subnet periodically and thus mobile does notneed to spend time in doing an ARP and wait after it obtains the IP address. I explain the details of theduplicate address detection mechanism in [33].

I infer from these experiments that the number of messages passing between the client and the serverduring IP address acquisition, end systems processing time and network load are some of the key factorsthat contribute to the delay. Proactive caching at the client, router or server assisted addres detectiontechnique help expedite the layer 3 address acquisition by reducing the time associated with duplicateaddress detection.

7.1.3 Limit binding update and redirect media

In this section, I describe several key optimization techniques that help reduce the transient data lossduring a mobile’s reconnection to a new PoA (Point of Attachment). I also demonstrate the validationof some of these following optimization techniques by means of experiments.

1. Limit the binding update within a domain

2. Redirect transient data from the previous network

3. Timebound multicasting

I have validated these techniques by experimenting with both application layer and network layermobility protocols. Details of my experimental analysis for the above two cases are provided in [63],[12]respectively. However, I only elaborate the reference [63] and discuss how I have experimented withthese techniques to achieve fast-handoff for an application layer mobility protocol such as SIP.

In case of intra-domain mobility, each visited domain may consist of several subnets. Every move toa new subnet within a domain causes the MH to send a re-INVITE to the CH containing its new care-ofaddress. If the re-INVITE request gets delayed due to path length or congestion, transient media packetswill continue to be directed to the old address. I assume that the visited network has an outbound proxy. Ienhance this proxy with the ability to temporarily register visitors [64]. The visitor obtains a temporary,

19

random identity from the visited network and uses it as its new address-of-record to register with theregistrar in the visited network. The MH informs the home registrar of this temporary address. It thenonly updates that registration with its current local IP address. This speeds up registration, but does notaddress the “delayed binding update” issue using SIP’s re-INVITE feature.

These proposed methods help alleviate transient data loss related to continuous hand-offs withina domain thus minimizing the effect of delay contributed during application layer rebinding as shownin Figure 5. I make of use certain SIP entities such as SIP registrar and RTP translator or NAT, theoutbound proxy and B2BUA as a mobility agent to implement these mechanisms. In-transit packets canbe redirected to a unicast or multicast address based on the movement pattern of the mobiles and usagescenario. I provide more details about the fast-handoff mechanism in [65].

Redirection of transient traffic from the network: I have implemented this mechanism using two ofthe approaches. In the first approach, each subnet within a domain is equipped with a redirect agent suchas RTP translator [66] that provides application-layer forwarding of RTP packets for a given addressand UDP port to a given network destination. The visited-network registrar described earlier receivesthe registration updates from the MH that has just moved, and immediately sends a request to the RTPtranslator in the network that the MH just left. The request causes the RTP translator to bind to theold IP address used by the MH and forward any incoming packets to the new address of the MH. Aftera set interval or after no media packets have been received by the RTP translator, the RTP translatorrelinquishes this old address and removes the forwarding table entry, assuming that the re-INVITE hasreached the CH.

The second approach uses a SIP outbound proxy. SIP requests typically traverse a SIP proxy in thevisited network, the outbound proxy. This outbound proxy can also support fast handoff, by using thedata in the MH-to-CH re-INVITE to configure the RTP translator or NAT. The advantage of this approachis that the outbound proxy usually has access to the Session Description Protocol (SDP) informationcontaining the MH media address and port, thus simplifying the configuration of the translator or NAT.On the other hand, this outbound proxy has to remember the INVITE information for an unboundedamount of time and become call stateful, since it needs the old information when a new re-INVITE isissued by the MH.

Limiting binding update traversal: I have demonstrated fast-handoff by introducing a signaling en-tity in the network that helps limit the binding update. In this case I use a back-to-back SIP user agent(B2BUA). A B2BUA consists of two SIP user agents where one user agent receives a SIP request, pos-sibly transforms it and then has the other part of the B2BUA re-issue the request. A B2BUA in eachdomain needs to be addressed by the MH in the visited domain. The B2BUA issues a new request tothe CH containing its own address as the media destination and then forwards the packets, via RTPtranslation or NAT, to the MH.

Timebound multicasting: Locally scoped multicast technique also helps avoid packet loss if the MHcan predict that it is about to move to a new subnet shortly. I have implemented this mechanism for bothapplication layer and network layer mobility management. In that case, it informs the visited registraror B2BUA of a temporary multicast address as its contact or media address. Once the MH has arrivedin its new subnet, it updates the registrar or B2BUA with its new unicast address, while continuing tolisten to multicast address. The use of scoped multicast is only effective if the MH can quickly acquirea multicast address and there is an inherent multicast infrastructure available.

I describe the performance evaluation of method 1. As shown in Figure 10, RT1, RT2 and RT3 areRTP translators in the respective subnets. These RTP translators forward the traffic associated with oneIP address and port number pair to another IP address and port number pair. RTP translator in eachof these subnets intercepts the traffic meant for the mobile host and sends it to the new address of themobile host after capturing it. This can be achieved by a combination of SIP-CGI and SIP Register[67]. Signal re-INVITE was delayed to simulate the network congestion or distance between CH and

20

MH CHVisited SIP

Registrar

Media

RT1 RT2 RT3

IP1

IP2(New Address)

Re-INVITE (2)

REGISTER 2’

Forwardtraffic(IP1:p1 ---> IP2:p1)

New traffic

IP3(New Address)

Re-INVITE

Re-REGISTER

Forwardtraffic(IP2:p1 ---> IP3:p1)

Transient Traffic dur ing the move

(1)

SIP-CGI (3)

Transient Traffic dur ing the move

SIP-CGI

First move

Second move

IP2

DelayBox

ServerPackets gain for SIP optimized handoff

0

10

20

30

40

50

60

0 20 40 60 80 100 120 140 160

Distance between CH and MH (Hops)

Nu

mb

er

of

pa

ck

ets

Packets gain for SIPoptimized handoff

Figure 10: Mobility optimization with media redirection

MH. Both VIC and RAT tools [68] were used to measure the performance of audio and video streamingtraffic respectively. I have experimented with two methods such as rtptrans and NAT-based iptables tohelp direct the transient traffic from the previous subnet to the new one. I tried few experiments withre-INVITE signals being delayed by 100 ms, 200 ms, 500 ms, 1 sec, 2 sec and 3 sec to show howRTP translator helps the media redirection and mitigates the packet loss during mobile’s movement. Imeasured the packet forwarding delay due to redirection at the registrar to be less than 1 ms when theiptables-based NAT approach was used, where as the RTPtrans approach added 4 ms of delay. Evenif the effect of binding update delay is minimized by redirection of packets, this mechanism by itselfcannot reduce the effect of layer 2 delay and IP address acquisition delay. Proactive mechanism anddynamic buffering techniques discussed in Section 7.2.2 will mitigate the effect of layer 2 and layer 3delay as well. Figure 10 compares the effect of SIP-optimized handoff compared to regular SIP-basedhandoff. I am currently implementing the B2BUA and multicast assisted handoff mechanisms.

7.1.4 Maintain security binding

I this section, I validate the mechanism that provides optimization by maintaining the security associationduring the handoff. The key principle introduced by this technique is to maintain the security associationof the end client even when the end-point identifier is changing. This avoids the delay due to re-bindingof security association at different layers during handoff.

I have designed and experimented a secured multi-interface mobility management scheme where amobile with dual interface moves between an enterprise network (e.g., 802.11), cellular network (e.g.,CDMA, GPRS) and hotspot (e.g.,802.11). By introducing an anchor point in the network I can achievesecured seamless roaming support without the need to tear down IPSec tunnels during each subnet move.I describe the details of the implementation and experimental analysis in [69] that use both mobile IPand SIP-based approaches. I have experimented with both CBR traffic (audio) and VBR traffic (video)and have analyzed the packet loss, delay and inter-packet gap during the handoff. In the absence ofsuch an optimized scheme, the mobile will be subjected to packet loss due to the delay associated withIPSEC tunnel setup and tear-down every time it moves. Thus in normal circumstances without anyoptimization, I observed that it takes about 10 seconds for the PPP negotiation to complete, 3 secondsfor DHCP address acquisition, about 300 ms for the x-MIP registration to complete, about 6 sec forVPN tunnel setup, 400 ms for the i-MIP registration, and 200 ms for mobile IP de-registration whenthe mobile is back. This results in degradation of real-time service. This experiment was done inWindows environment and thus provides a different value for PPP setup than that described in Section7.2.1. However deploying a combination of optimization technique that helps maintain the securitybinding and introduces a make-before-break technique, we were able to obtain zero packet loss during

21

the handover from 802.11 networks to cellular and back. Figure 11 shows the results of the experimentinvolving secured mobility across heterogeneous networks for both optimized and non-optimized cases.Low gradient in the graph implies the low network speed within a cellular network. Although there wasno packet loss in the optimized case, the mobile received few out of order packets during its movementfrom cellular network to 802.11 network because of the transient packets in the path that arrived in theWAN interface at a later point.

Non-Make-before-break

31000

32000

33000

34000

35000

0 50 100 150 200 250 300

Time in Seconds

RT

P S

eq

ue

nc

e

RTP Sequence

802.11b

Cellular

802.11bPacketLoss

Packet loss

802.11-Cellular Secured Handoff

180019002000210022002300240025002600

0 20 40 60 80 100 120 140 160 180

Time in SecondsR

TP

Pa

ck

et

Se

qu

en

ce

RTP sequence duringhandoff

No packet lossOut-of-order-packet

802.11b

Cellular

802.11b

Figure 11: Effect of security rebinding and its optimization

7.2 Proactive methodologies

In this section, I describe and demonstrate several proactive methodologies that can provide mobilityoptimization for interactive traffic.

7.2.1 Make-before-break technique

In this section, I demonstrate the validation of make-before-break technique by way of experimentalresults.

I have designed and demonstrated a make-before-break technique that provides mobility optimiza-tion for heterogeneous access networks. In order to reduce the packet loss due to delay in layer 2 andlayer 3 configuration I have implemented a make-before-break algorithm that sets up the layer 3 config-uration in the new network using the new interface while the mobile still keeps on communicating usingthe current access network.

I have validated this mechanism by experimenting with both network layer protocol such as mo-bile IP and application layer protocol such as SIP-based mobility and have used heterogeneous accesstechnologies such as 802.11 and CDMA1XRTT. From the experimental analysis it is found that in theabsence of any make-before-break technique, total amount of delay during handover consists of the se-quential delays due to discovery of new network point of attachment, setting up layer 2 and layer 3association and finally sending the binding update to the correspondent node. I measured the delay andpacket loss during the movement from 802.11 network to CDMA network for normal handoff case andcompared it with optimized case. In normal experimental set up, involving both MIP and SIP, time takenfor PPP setup is about 16 seconds. MIP re-registration over PPP interface took about 540 ms. I observedthat the mobile lost about 411 packets during this handoff due to the delay. In case of SIP-based mobilityI observed similar delay for PPP setup, however binding update was sent directly to the correspondenthost. A complete binding update process took about 900 ms due to application layer processing on theend hosts and have resulted in several re-transmission. Some of these results of unoptimized handoff areshown in Figure 1 of Section 3.

22

However, by using make-before-break technique I did not observe any packet loss for both MIPand SIP, although I observed initial jitter during the handover from 802.11 to CDMA network andout-of-order packet during handover from CDMA to 802.11 networks. I have described a completeexperimental handoff analysis of this work in [70].

7.2.2 Media independent proactive handover

In this section, I describe the following few optimization techniques that constitute media independentproactive mechanism framework. I also validate some of these mechanisms by conducting an experimen-tal analysis of the proactive model using both network layer and application layer mobility managementprotocols such as MIPv6 and SIP-based mobility protocol.

1. Proactive discovery of neighboring networks

2. Pre-authentication

3. Proactive security association

4. Pre-configuration

5. Proactive binding update

6. Dynamic buffering

I have developed Media independent Pre-Authentication (MPA) [71] framework that is a mobile-assisted, secure handover optimization scheme and works over any link-layer and with any mobilitymanagement protocol. With MPA, a mobile node is not only able to securely obtain an IP address andother configuration parameters from a candidate target network, but also is able to send and receiveIP packets using the obtained IP address and other configuration parameters, before it attaches to thecandidate target network. Thus it makes it possible for the mobile node to complete the binding updateand use the new care-of address before performing a handover at link-layer. MPA scheme providesthe optimization techniques by taking care of the key mobility components such as network discovery,authentication, configuration, security association, binding update proactively. In addition, it enhancesthe optimization technique by providing dynamic buffering and copy-and-forward mechanism at theedge of the network. I explain briefly some of the initial results of the MPA scheme.

Proactive network discovery: Here, I have focused my work on network discovery methods, wherethe client discovers the neighboring elements (e.g., routers, DHCP servers, SIP servers) and commu-nicates with these entities before it actually moves into these networks. This helps expedite the au-thentication and IP address acquisition part of the handoff process. As part of the initial work, I haveimplemented an information server using RDF (Resource Description Framework) and XML schemathat helps to discover the networks and the resources. I describe the details of how we can take advan-tage of this discovery scheme for fast-handoff in reference [72].

Pre-authentication: As described in Section 5, authentication and security association are two im-portant steps that need to be completed before the client reestablishes the existing session. By meansof proactive discovery process a mobile can discover the authentication servers in the target networks.A successful pre-authentication can also give rise to pre-configuration and estbalishment of security as-sociation between the mobile and network elements. As part of the initial results, I have validated thepre-authentication techniques using PANA protocol and have experimented with both application layerand network layer mobility protocols.

23

Proactive security association: By using the security association established between the mobile nodeand the authentication agent in a candidate target network during the pre-authentication phase, it ispossible to bootstrap link-layer security and network layer security in the candidate target network whilethe mobile node is in the current network. After the mobile node chooses a specific candidate network asthe target network and switches to a new point of attachment in the target network (which now becomesthe new network for the mobile node), it executes a secure association protocol such as IEEE 802.11i4-way handshake using the PMK (Pair-wise Master Key) in order to establish PTKs (Pair-wise TransientKeys) and GTKs (Group Transient Keys) used for protecting link-layer packets between the mobile nodeand the point of attachment. No additional execution of EAP authentication is needed here. Similarly,a network layer security association can also be established in the access router of the target network.This proactive procedure reduces the associated delay due to security association and authentication.

Pre-configuration: In the following paragraph I describe different components associated with Pre-configuration phase.

Proactive address acquisition: As discussd in the Section 7.1, in general, IP address acquisitionand configuration process take of the order of few hundred milliseconds to few seconds depending uponthe type of IP address acquisition process and operating systems of the clients and the servers. SinceIP address acquisition is part of the handover process, it adds to the handover delay. In case of MPAassisted handoff, I propose to obtain the IP address of the target network while in the current networkthus reducing the handoff delay contributed by the IP address acquisition process. I have demonstratedproactive address acquisition for both IPv4 and IPv6 networks using the PANA-based authenticationprotocol. I have described the details in [71].

Proactive duplicate address detection: Duplicate IP address detection mechanism helps maintainthe uniqueness of the address obtained. In such scenario, the client usually does a duplicate addressdetection based on ARP (Address Resolution Protocol) or IPv6 Neighbor Discovery before assigningthe IP address. This detection procedure may take up to 4 sec to 15 sec [73] and thus contributes to alarger handover delay. As part of proactive IP address acquisition process, this detection is performedahead of time while the mobile is in the previous network. It thus reduces the overall handover delayfactor associated with L3 configuration.

Proactive address resolution: Address resolution is the process of mapping the network addresswith the layer 2 address so that packets get delivered to the right node. During the process of pre-configuration the mobile can also install the ARP mapping of the neighboring first hop router withwhom it will need to communicate after attaching to the new network. Having the prior knowledge ofARP, both the neighboring first hop router and the mobile do not spend time in discovering each other atlayer 2.

Proactive mobility binding update: In Section 7.1.3, I have proposed ways of reducing the effectof delay and packet loss due to binding update, that usually takes place after the mobile has movedto the new network. In the proposed MPA scheme, it is done proactively. This is achieved by settingup the tunnel between the mobile and the next hop router and sending the binding update proactivelywith the new cached IP address that it would obtain in the new network. This effectively removes theneed for sending a binding update after the mobile moves to the new network and thus reduces thedelay associated with the binding update. This optimization becomes more prominent when the distancebetween CH and MH is large.

Dynamic buffering: Although IP address acquisition and binding update are optimized in case ofproactive handover, there may be some transient packets that can be lost during link-layer handover and

24

BufferingEnabled

BufferingDisabled

BufferingEnabled+ ROEnabled

Buffering Disabled+ RO Enabled

Buffering Enabled+ RODisabled

Buffering Disabled+ RO Disabled

HandoffParameters

3.00n/a3.00n/a2.00 n/aAvg. Buffered Packets

20.00n/a50.00n/a50.00n/aBuffering period (ms)

13.00n/a50.60n/a29.33 n/aAvg. packet jitter (ms)

29.00n/a66.60n/a45.33n/aAvg. inter-packet arrival time during handover (ms)

16.0016.0016.0016.0016.0016.00Avg. inter-packet interval (ms)

01.5000.660 1.33Avg. packet loss

5.004.004.004.004.334.00L2 handoff (ms)

MIPv6Mobility Type SIP Mobility

Table 6: Results from proactive handoff

until the traffic is directed to the mobile node after attaching to the target network. Bicasting or bufferingthe transient packets at the access router can be used to minimize or eliminate the packet loss. However,bicasting does not eliminate packet loss if link-layer handover is not seamlessly performed. In addition,buffering in the network node introduces an additional end-to-end delay in the packet delivery of thein-handoff packets, but mitigates the effect of handoff by reducing the packet loss. While this additionalend-to-end delay may not affect the streaming traffic, interactive traffic such as VoIP application cannottolerate the large delay jitter. I have implemented a dynamic buffering technique [74] that ensureszero packet loss while introducing additional end-to-end delay that is well within threshold. I haveexperimented with two kinds of buffering scheme such as time limited buffering and explicit signalingbuffering techniques to help the proactive handover. But these techniques can also be applied to reactivehandover scenario.

I have implemented MPA framework to validate all of the above techniques using network layerand application layer mobility protocols such as MIPv6 and SIP respectively. Table 6 shows the resultsof MPA-based experiment with pre-configuration, pre-authentication and dynamic buffering. Resultsdemonstrate how a combination of proactive handoff mechanisms can help reduce the packet loss anddelay to an acceptable threshold level.

7.3 Multi-layer methodologies

In this section, I describe the experimental details of the optimization technique that takes advantageof mobility protocols at different layers and uses a combination of micro and macro mobility protocolsbased on the mobile’s movement pattern.

Mobility management protocols at each layer are best suited to work for a specific type of applicationand movement pattern of the mobile. For example an application layer mobility management may besuited to work for interactive traffic such as VoIP, while a local mobility management protocol may workwell while a mobile’s movement is confined to intra-domain. A network layer mobility managementprotocol works well to support TCP-based application. In such a situation a policy-based approachsupporting a multilayered mobility management will provide some level of optimization compared tothe case when only one mobility protocol is used for both global mobility and local mobility.

I have designed and prototyped a multi-layer mobility management scheme to validate this opti-mization approach. It consists of three mobility protocols such as SIP-based, MIP-LR and MMP (MicroMobility Protocol) that work in conjunction with a policy manager and executes the desired mobilityprotocol based on the type of application and mobile’s movement pattern. SIP-based mobility man-agement is used for real-time communication, and application layer MIP-LR is used for non-real-timetraffic during a node’s movement between two different domains while MMP takes care of the move-ment within a domain. I describe the complete implementation of the optimized multi-layer mobilitymanagement scheme in [75].

25

1 2 3 4

gateway

BS’s

MH

Host BasedRoutingMicro-MobilityDomain

CH

1 2 3 4

gateway

BS’s

MH

Host BasedRoutingMicro-MobilityDomain(Cellular IP)

CHMacro Mobility (SIP, MIP)

0

5

10

15

20

25

30

0 0.05 0.1 0.15 0.2 0.25 0.3Link Latency (s)

Th

rou

gh

pu

t (k

by

te/s

) MIP 1MIP 0.2MIP 0.1HBR SS 1HBR SS 0.2HBR SS 0.1

Figure 12: Multilayer mobility management scenario and results

MMP is a derivative of Cellular IP and shares certain benefits of host based routing and forwarding-cache-based techniques exploiting hierarchical structure of military networks. MMP is designed as amicro-mobility protocol to handle intra-domain mobility and works in conjunction with SIP and MIP-LR, where SIP and MIP-LR handle macro-mobility and take care of inter-domain movement. Results ofsimulation and experimental results for MMP are explained in [76].

To support real-time communication during the mobile’s movement between the domains, the mobilesends a re-INVITE to the CH to keep the session active, similarly a MIP-LR UPDATE message is sentto CH for the TCP/IP traffic. But for any subsequent move within the new domain re-INVITE or updatemessages are not sent, since MMP takes care of routing the packets properly within that domain. Thishelps limiting the binding update from traversing a long distance. More details of this scheme can befound in reference [77].

Figure 12 illustrates a typical scenario where a combination of macro-mobility and micro-mobilityprotocols can work together. The results shown in Figure 12 demonstrate how using a micromobilityprotocol is more optimized when the movement of the mobile is confined to a domain. Thus use of theright mobility protocol based on the mobile’s movement pattern and application type provides the mostoptimized solution. Most recently the IETF is considering a network controlled mobility protocol tosupport mobile’s movement within a domain.

7.4 Methodologies for simultaneous mobility

Simultaneous mobility problem occurs when two mobile nodes that are part of a communication session,and they both move such that the binding updates that they send to each other are both lost throughbelated arrival binding update, and such that the communication session never returns from interruptedstate to normal state and is ended.

In this section, I analyze the problem associated with simultaneous mobility and describe some of theoptimization techniques that can mitigate this problem. These techniques can broadly be classified intotwo: sender-based, receiver-based mechanisms. Retransmission, forwarding, redirecting, proactive-forwarding and proactive-redirecting are some of the mechanisms that have been analyzed for each ofthese mobility protocols.

If the probability that any particular handoff suffers from simultaneous mobility problem be ��� ,and the probability that at least one out of N handoffs in a given session suffers from the simultaneousmobility problem be ��� . Thus ��� = 1 ��� - ��� ) ���� ���� P � = E[ � + � ]/ � . Where � and � are the amount oftime needed for a binding update to reach from A to B and vice-versa and � is the average inter-handofftime. Figure 13 shows how the simultaneous mobility problem is affected by a combination of latency

26

and inter-handoff time. As part of initial results, I have conducted a preliminary analysis of simultaneousmobility of IP hosts using SIP, MIPv6 and MIP-LR based mobility protocols and have designed theoptimization techniques that help reduce the effect of simultaneous mobility problem. Details of thesemechanisms and results can be found in references [78], [79].

Figure 13: Probability for handoff with simultaneous mobility

7.5 Methodologies for multicast streaming

In this section, I introduce a hierarchical scope-based multicast streaming architecture, describe the mo-bility optimization techniques for multicast streaming and validate these techniques with experimentalresults.

The very process of joining or leaving a specific multicast group while changing the cell or subnetis similar to surfing a TV or Radio and flipping the channels [80]. Unlike unicast traffic, the multicastcommunication is receiver initiated. Thus triggering techniques play an important role for efficientmultimedia stream delivery. In order to maintain minimum loss and latency during the client’s movementit is desirable to minimize the handoff time and to provide almost instantaneous flow of multicast streamby adopting faster triggering techniques.

Latency associated with receiving continuous multicast stream from a single source during the mo-bile’s movement depends upon several components such as detection of a new cell, subnet, time totrigger, time needed for the router to join the upstream router and actual media delivery. Unlike unicasttraffic, layer 3 configuration time on the client does not affect the multicast stream delivery.

Traditionally, triggering delay associated with a mobile’s ability to join a multicast tree after a mobilehas moved, depends on Internet Group Management Protocol (IGMP) [81] membership query report. Atypical query interval for the IGMP is by default 125 seconds [82], although this value is configurable.Flament et al [83] show that by using IGMP, a host will wait for 65 seconds on average before it cancontinue to receive the multicast traffic after the handover. Thus, IGMP in its current form is not suitableto support real time communication because of the associated triggering delay. In case of a client’smovement between access points within a subnet, CGMP (Cisco Group Management Protocol) or IGMPsnooping [82] take care triggering the multicast stream.

I have designed and prototyped a hierarchical scope-based multicast content distribution networkcalled MarconiNet [84]. This architecture uses the IETF protocols, SDP [85], SAP [86], SIP [19] andtakes advantage of the real-time feedback signal RTCP to provide many flexible features such local-ized advertisement, news broadcast, location specific information, QoS guarantee and optimized intra-domain handoff for the mobile users. In this architecture, the higher level of the two, global multicastexists between the broadcasting stations (RSCs) and the local stations (RASs). At the lower level of thehierarchy, a locally scoped multicast session is created for each broadcasting station between the serverand the listening clients (IMC) that can be privately scoped. Local server interacts with the commercialserver to provide stream control using protocols such as SIP and RTSP [87]. I have prototyped differ-

27

ent functional components of the architecture such as local and global program management, channelestimation, application layer triggering and handoff involving multiple servers. However, I focus myanalysis on fast-handoff approaches only.

In the following paragraphs, I describe several optimization techniques that can provide faster mul-ticast stream delivery.

Reactive triggering: Normally, when a client moves to a new subnet, it sends the join query via IGMPreport. In some cases however IGMP could be modified to provide aggregate group report to reduce thejoin latency [88]. In this work, I have implemented an application layer triggering mechanism based onRTCP to facilitate the join and leave processes. Triggering at the lower hierarchy is accomplished usingRTCP feedback signal, but the local server triggers the upstream router using IGMP method. Using anRTCP-based triggering technique offers a solution at the user space compared to IGMP which is networklayer.

Alternatively, group membership information can also be passed during the client’s configuration inthe new network. During a node’s movement between subnets it can send the request for the previouslysubscribed multicast address as part of DHCP discover message or PPP’s IPCP option. During the pro-cess of obtaining the IP address from the DHCP server, the client can send the unsolicited “JOIN” requestfor the desired locally scoped multicast address to the server. This process allows the JOIN procedureduring client’s configuration itself and provides the optimization using parallelization technique.

Proactive triggering: Proactive join method reduces the join latency for an impending client at theexpense of extra flow of stream in the adjacent cell for certain duration. I describe two kinds of proactivejoin schemes below.

First approach proposes deploying proxy agents in each subnet. These proxy agents join the up-stream multicast tree on behalf of the local servers even before the clients move into the new cell. Thelocal server can signal the proxy agents to notify the proxy agents about the impending host’s alreadysubscribed multicast address. Thus multicast proxy sends the IGMP query messages beforehand onbehalf of the local servers and can help forward the global stream to the respective global multicast ad-dresses (e.g., for areas where these clients are impending to move) in each subnet for a specific periodof time determined by the client entering to the cell.

The second approach does not include the proxy agent. Rather for each of neighboring stationssharing an overlapping area with another station there is an associated multicast announcement (address).Each local server can subscribe to this address and find out the group address that impending clientis subscribed to. Just before a mobile node leaves (decides to leave) the cell, the mobile sends themovement imminent signal to the local announcement address and the currently subscribed address.The local server in turn announces that to the shared multicast addresses that the neighboring localstations subscribe to in the global space. In the absence of this association, the neighboring server sendsan IGMP message to the upstream router and redirects the stream to the local cells even before the clienthas moved to the new cell. This helps minimizing the interruption.

Figure 14 shows an experimental environment where the fast-handoff experiment for multicaststreaming is carried out. � , ��� , � � are the globally routable subnets connected to the primary interfacesof the local servers S1, S2 and S3, where as � , ��� , � � are the local subnets connected to the secondary in-terfaces of the servers where the mobile resides. I have conducted fast-handoff experiments for cell andsubnet mobility using multimedia applications such as rat, vic [68] and took several measurements in-cluding the time for movement detection, IP address acquisition, join and leave latency by using networklayer IGMP, CGMP (Cisco Group Management Protocol) and application layer RTCP signaling.

Figure 15 (a) shows the protocol sequence during the subnet handoff in 802.11 environment. Itexposes the associated join latency during a normal handoff without any optimization. However asshown in Figure 15 (b), proxy-based expedited join reduces the join latency to almost zero.

28

R

R1 R2

S1 S2 S3 S4

Multicast Switch

S5

Ia Ib Ic Id

M1

m1

ia ib ic id

Mx

Lmx

ib

MulticastAgent

Mx

MsxSharing multicast address

Lmxn1

M1

m1Same subnetLocal

Ad

M1

LocalAdlm1

m1 lm1

Multicastproxy

Multicast Stream Server

Figure 14: Proxy-based fast-handoff for multicast streaming

Figure 15: Join latency and its optimization in 802.11 environment

8 Roadmap for Future Work

In this section, I provide a roadmap for the completion of my research and describe briefly the additionalwork that I will perform.

8.1 Validation of mobility optimization methodologies using Petri net models

I will complete the formulation of the formal systems model that can represent a mobility event andcharacterize the mobility optimization methodologies based on the common systems properties. I willuse specific modeling tool such as Timed Petri net to validate the optimization methodologies of anymobility protocol. I plan to derive the optimized systems model for several key mobility optimizationmethodologies. I will apply different Petri net reduction rules such as serial fusion, pre-fusion or lateralfusion to derive the optimized model and compare its efficiency with the non-optimized standard mo-bility model. I plan to use the package TimeNet [89] to conduct the performance analysis. Expectedcompletion date is July 2006.

8.2 Analytic comparison of SIP-based mobility and MIPv6

Mobility approaches adopted by Mobile IPv6 and SIP-based mobility protocols exhibit a lot of similari-ties in many of the required functions such as binding update, location update to support a mobility event.As part of an initial set of results [62] I have experimented with both SIP based mobility scheme for IPv6and MIPv6 in the testbed. I will carry out an analytical and performance comparison between both ofthese mobility approaches using different mobility parameters such as mobility rate, cell residence time,packet-to-mobility ratio etc. Expected completion date is August 2006.

29

8.3 Comparison of MPA with other fast-handoff mechanism such as FMIPv6

As part of the initial work, I have demonstrated the experimental validation of MPA mechanism us-ing two mobility protocols such as SIP and MIPv6. I will carry out a comparative analysis of MPAwith similar fast-handoff approaches such as proactive mode of MIPv6 called FMIPv6 [90]. Expectedcompletion date is September 2006.

8.4 Analysis of ping-pong effect during proactive handover

During proactive handover, ping-pong effect is a common phenomenon. Ping-pong effect affects the op-eration of any mobility protocol in terms of utilizing resources etc. I plan to study the effect of MPA op-timization under ping-pong situation and will develop a handoff selection algorithm that could decreasethe probability of ping-pong effect during MPA operation. Expected completion date is October 2006.

8.5 Summary of plan for completion of research

Table 7 shows my plan for completion of the research. Thus, I plan to defend my thesis in April 2007.

Timeline Work ProgressMobility systems analysis completedPreliminary mobility systems modeling completedMobility optimization methodologies completedPrototyping and experimental verification of mobility optimization methodologies completed

Jul. 2006 Validation of mobility optimization methodologies using Petri net model ongoingAug. 2006 Comparative analysis of network layer and application layer mobility protocols ongoingSep. 2006 Comparison of MPA with FastMIPv6 ongoingOct. 2006 Analysis of ping-pong effect during proactive handover ongoingJan. 2007 Thesis writingApr. 2007 Thesis defense

Table 7: Plan for completion of my research

9 Conclusion

This thesis proposal contributes to the general theory of optimized handover and has addressed the needfor a formal systems model that can characterize a mobility event and the associated formal mobilityoptimization methodologies. It provides a systematic and formal approach to a mobility event. Aftera thorough analysis of the abstract properties associated with any mobility protocol (e.g., cellular, IP),it represents these in the form of a set of communicating finite state machines and proposes a formalsystems model using a Deterministic Timed Petri net. This model characterizes the common optimiza-tion methodologies that can act as the design principles for any new mobility protocol as well as canhelp evaluate the effectiveness of a mobility protocol. Initial series of works have resulted in formula-tion of a formal systems mobility model, identification of the optimization methodologies and validationof the associated optimization techniques by means of analysis, simulation and experiment. Validationincludes several types of mobility events such as heterogeneous handover, simultaneous mobility andmulticast mobility. Roadmap and timeline to completion of the research has been laid out. These in-clude comparative analysis of candidate mobility protocols, their optimization techniques and validationof the optimization methodologies by means of Petri net models and comparing them with experimentalresults.

30

References

[1] N. Tripathi, J. Reed, and H. F. VanLandingham, “Handoff in cellular systems,” IEEE PersonalCommunications Magazine, vol. 5, Dec. 1998.

[2] C. E. Perkins, “IP mobility support for IPv4,” RFC 3344, Internet Engineering Task Force, Aug.2002.

[3] D. B. Johnson, C. E. Perkins, and J. Arkko, “Mobility support in IPv6,” RFC 3775, Internet Engi-neering Task Force, June 2004.

[4] A. Dutta, O. Altintas, W. Chen, and H. Schulzrinne, “Mobility approaches for all IP wirelessnetworks,” in SCI, (Orlando, Florida), July 2002.

[5] A. Dutta, F. Vakil, J. Chen, M. Tauil, S. Baba, and H. Schulzrinne, “Application layer mobilitymanagement scheme for wireless Internet,” in 3Gwireless, (San Francisco), p. 7, May 2001.

[6] C. E. Perkins, “IP encapsulation within IP,” RFC 2003, Internet Engineering Task Force, Oct. 1996.

[7] C. E. Perkins, “Minimal encapsulation within IP,” RFC 2004, Internet Engineering Task Force,Oct. 1996.

[8] S. K. Park, “Rapid commit option for DHCPv4,” Internet Draft draft-ietf-dhc-rapid-commit-opt-00,Internet Engineering Task Force, Dec. 2003. Work in progress.

[9] R. Jain, T. Raleigh, D. Yang, L. F. Chang, C. J. Graff, M. Bereschinsky, and M. Patel, “Enhancingsurvivability of mobile Internet access using mobile IP with location registers,” in Proceedings ofthe Conference on Computer Communications (IEEE Infocom), (New York), Mar. 1999.

[10] A. T. Campbell, J. Gomez, S. Kim, A. G. Valk, C.-Y. Wan, and Z. R. Turnyi, “Design, implementa-tion, and evaluation of cellular IP,” IEEE Personal Communications Magazine, vol. 7, pp. 42–49,Aug. 2000.

[11] R. Ramjee, T. F. LaPorta, L. Salgarelli, S. Thuel, K. Varadhan, and L. E. Li, “IP-based accessnetwork infrastructure for next-generation wireless networks,” IEEE Personal CommunicationsMagazine, vol. 7, pp. 34–41, Aug. 2000.

[12] S. Das, A. Dutta, A. McAuley, A. Misra, and S. Das, “IDMP: an intra-domain mobility manage-ment protocol for next generation, wireless networks,” IEEE Personal Communications Magazine,June 2002.

[13] S. Thomson and T. Narten, “IPv6 stateless address autoconfiguration,” RFC 2462, Internet Engi-neering Task Force, Dec. 1998.

[14] R. E. Droms, I. W. Ed., J. Bound, B. Volz, T. Lemon, C. E. Perkins, and M. Carney, “Dynamic hostconfiguration protocol for IPv6 (dhcpv6),” RFC 3315, Internet Engineering Task Force, July 2003.

[15] A. C. Snoeren and H. Balakrishnan, “An end-to-end approach to host mobility,” in ACM/IEEEInternational Conference on Mobile Computing and Networking (MobiCom), (Boston, Mas-sachusetts, USA), pp. 155–166, Aug. 2000.

[16] D. A. Maltz and P. Bhagwat, “MSOCKS: an architecture for transport layer mobility,” in Proceed-ings of the Conference on Computer Communications (IEEE Infocom), (San Francisco, California),p. 1037, March/April 1998.

[17] M. Leech, M. Ganis, Y. Lee, R. Kuris, D. Koblas, and L. Jones, “SOCKS protocol version 5,” RFC1928, Internet Engineering Task Force, Mar. 1996.

31

[18] S. N. Koh et al., “Use of SCTP for seamless handover,” internet draft, Internet Engineering TaskForce, Feb. 2003. Work in progress.

[19] J. Rosenberg, H. Schulzrinne, G. Camarillo, A. R. Johnston, J. Peterson, R. Sparks, M. Handley,and E. Schooler, “SIP: session initiation protocol,” RFC 3261, Internet Engineering Task Force,June 2002.

[20] H. Schulzrinne and E. Wedlund, “Application-layer mobility using SIP,” Mobile Computing andCommunications Review (MC2R), vol. 4, pp. 47–57, July 2000.

[21] T.Kivinen and H.Tschofenig, “Design of mobike protocol,” Internet Draft draft-ietf-mobike-design-08, Internet Engineering Task Force, Mar. 2006. Work in progress.

[22] R.Moskowizt, P.Nikander, P.Jokela, and T.Henderson, “Host identity protocol,” Internet Draftdraft-ietf-hip-base-05, Internet Engineering Task Force, Mar. 2006. Work in progress.

[23] P. Calhoun, G. Montenegro, and C. E. Perkins, “Mobile IPv4 regional registration,” Internet Draftdraft-ietf-mobileip-reg-tunnel-08, Internet Engineering Task Force, Nov. 2003. Work in progress.

[24] K. Malki, “Low latency handoffs in mobile IPv4,” Internet Draft draft-ietf-mobileip-lowlatency-handoffs-v4-08, Internet Engineering Task Force, Jan. 2004. Work in progress.

[25] C. E. Perkins and K.-Y. Wang, “Optimized smooth handoffs in mobile IP,” in Fourth IEEE Sympo-sium on Computers and Communications (ISCC ’99), (Red Sea, Egypt), pp. 340–346, July 1999.

[26] P. Calhoun, T. Hiller, J. Kempf, P. McCann, C. Pairla, A. Singh, and S. Thalanany, “Foreign agentassisted hand-off,” internet draft, Internet Engineering Task Force, Nov. 2000. Work in progress.

[27] H. Soliman, C. Castelluccia, K. Malki, and L. Bellier, “Hierarchical mobile IPv6 mobility manage-ment (hmipv6),” internet draft, Internet Engineering Task Force, July 2003. Work in progress.

[28] F. Vakil, D. Famolari, S. Baba, and D. Famolari, “Virtual soft hand-off in IP-Centric wirelessCDMA networks,” in 3G Wireless 2001, (San Francisco), Delson, Delson, May 2001.

[29] S. K. Park and Y. W. Han, “IPv6 DAD optimization goals and requirements,” Internet Draftdraft-park-dna-ipv6dadopt-requirement-02, Internet Engineering Task Force, Dec. 2003. Workin progress.

[30] Y. Han et al., “Advance duplicate address detection,” internet draft, Internet Engineering TaskForce, July 2003. Work in progress.

[31] A.Forte, S.Shin, and H.Schulzrinne, “Passive duplicate address detection for the dynamic hostconfiguration protocol for ipv4,” Internet Draft draft-forte-dhc-passive-01, Internet EngineeringTask Force, Mar. 2006. Work in progress.

[32] A. McAuley, S. Das, S. Madhani, S. Baba, and Y. Shobatake, “Dynamic registration and config-uration protocol (DRCP),” internet draft, Internet Engineering Task Force, July 2000. Work inprogress.

[33] IEEE, GPS assisted Fast-handoff Mechanism for Real-Time Communication, (Sarnoff Symposium,Princeton, NJ), IEEE, 2006.

[34] C. Williamson, T. Harrison, W. L. Mackrell, and R. B. Bunt, “Performance evaluation of the MoMmobile multicast protocol,” ACM Mobile Networks and Applications (MONET) Journal, vol. 3,pp. 189–201, Aug. 1998.

[35] C. Lin and K.-M. Wang, “Mobile multicast support in IP networks,” in Proceedings of the Confer-ence on Computer Communications (IEEE Infocom), (Tel Aviv, Israel), Mar. 2000.32

[36] I. Romdhani, M. Kellil, H.-Y. Lach, A. Bouabdallah, and H. Bettaher, “IP mobile multicast:Chalenges, solutions and open issues,” motorola technical report, Motorola Labs, Paris, France,July 2002.

[37] G. Xylomenos and G. C. Polyzos, “IP multicast for mobile hosts,” IEEE Communications Maga-zine, vol. 35, Jan. 1997.

[38] U. Varshney and S. Chatterji, “Architectural issues in IP multicasting over wireless networks,”in IEEE conference on Wireless Communication and Networking Conference, IEEE, IEEE, June1999.

[39] A. Acharya, A. Bakre, and B. Nath, “IP multicast extensions for mobile internetworking,” Techni-cal Report LCSR-TR-243, Department of Computer Science, Rutgers University, New Brunswick,New Jersey, Apr. 1995.

[40] J.-L. C. Wu, “An IP mobiity support architecture for 4GW wireless infrastructure,” in PCC, Nov.1999.

[41] A. McAuley, E. Bommaiah, A. Misra, R. Talpade, S. Thomson, and K. C. Young, “Mobile multicastproxy,” in IEEE Milcom, (Atlantic City, New Jersey), Nov. 1999.

[42] C. L. Tan and S. Pink, “Mobicast: A multicast scheme for wireless networks,” Mobile Networksand Applications, vol. 5, pp. 259–271, Jan. 2000.

[43] J. Mysore and V. Bharghavan, “A new multicasting-based architecture for Internet host mobility,”in ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom), (Hun-gary), Sept. 1997.

[44] International Workshop on Petrinet and Performance Models, On Petri Net-based ModelingParadigms for the Performance Analysis of Wireless Internet Access, IEEE, 2001.

[45] R. M. Amadio and S. Prasad, “Modelling IP mobility,” tech. rep., INRIA Sophia Antipolis, 1997.http://www.inria.fr/rrrt/rr-3301.html, accessed July 2002.

[46] R. Johnson, T. Palaniappan, and M. Stapp, “Goals of detecting network attachment in ipv6,” RFC4135, Internet Engineering Task Force, Aug. 2005.

[47] M. Georgiades, “Context transfer support for ip-based mobility management,” tech. rep., 2004.

[48] B. Aboba, L. Blunk, J. Vollbrecht, J. Carlson, and E. H. Levkowetz, “Extensible authenticationprotocol (EAP),” RFC 3748, Internet Engineering Task Force, June 2004.

[49] S. Blake-Wilson, M. Nystrom, D. Hopwood, J. Mikkelsen, and T. Wright, “Transport layer security(TLS) extensions,” RFC 3546, Internet Engineering Task Force, June 2003.

[50] T. Murata, “Petri nets: Properties, analysis and applications,” Proceedings of IEEE, Apr. 1989.

[51] R. Sloan and U. Buy, “Reduction rules for time petri nets,” Acta Informatica, 1996.

[52] W.M.Zuberek and W.Kubiak, “Timed petri nets in modeling and analysis of simple schedules formanufacturing cells,” Elsevier Science Journal, Computers and Mathematics with Applications,Aug. 1999.

[53] A. Dutta, P. Agrawal, J.-C. Chen, S. Das, D. Famolari, Y. Ohba, S. Baba, and A. McAuley, “Real-izing steaming wireless Internet telephony and streaming multimedia testbed,” Elsevier, Computerand Communication, Oct. 2003.

33

[54] E. Wedlund and H. Schulzrinne, “Mobility support using SIP,” in 2nd ACM/IEEE InternationalConference on Wireless and Mobile Multimedia (WoWMoM), (Seattle, Washington), Aug. 1999.

[55] D. Forsberg, “Protocol for carrying authentication for network access (PANA),” Internet Draftdraft-ietf-pana-pana-03, Internet Engineering Task Force, Feb. 2004. Work in progress.

[56] P. Calhoun, J. Loughney, E. Guttman, G. Zorn, and J. Arkko, “Diameter base protocol,” RFC 3588,Internet Engineering Task Force, Sept. 2003.

[57] M. Baugher et al., “The secure real-time transport protocol,” internet draft, Internet EngineeringTask Force, July 2003. Work in progress.

[58] S. Dusse, P. Hoffman, B. Ramsdell, L. Lundblade, and L. Repka, “S/MIME version 2 messagespecification,” RFC 2311, Internet Engineering Task Force, Mar. 1998.

[59] A. Dutta, S. Das, P. Li, A. McAuley, Y. Ohba, S. Baba, and H. Schulzrinne, “Secured mobilemultimedia communication for wireless Internet,” in International Conference on Network Sensingand Control, (Taipei, Taiwan), IEEE, IEEE, Mar. 2004.

[60] A. J. Tuominen and H. L. Petander, “MIPL mobile IPv6 for linux in HUT campus network medi-apoli,” in Proceedings of Ottawa Linux Symposium, 2001, (Ottawa, Canda), June 2001.

[61] T. Narten, E. Nordmark, and W. Simpson, “Neighbor discovery for IP version 6 (ipv6),” RFC 2461,Internet Engineering Task Force, Dec. 1998.

[62] N. Nakajima, A. Dutta, S. Das, and H. Schulzrinne, “Handoff delay analysis and measurement forSIP based mobility in IPv6,” in Conference Record of the International Conference on Communi-cations (ICC), May 2003.

[63] A. Dutta, S. Madhani, W. Chen, O. Altintas, and H. Schulzrinne, “Fast-handoff schemes for appli-cation layer mobility management,,” in IEEE PIMRC, IEEE, IEEE, Sept. 2004.

[64] H. Schulzrinne, “SIP registration,” internet draft, Internet Engineering Task Force, Apr. 2001. Workin progress.

[65] A. Dutta, H. Schulzrinne, S. Madhani, O. Altintas, and W. Chen, “Optimized fast-handoff schemesfor application layer mobility management,” Mobile Computing and Communications Review(MC2R), vol. Volume 6, Nov. 2002.

[66] H. Schulzrinne, S. Casner, R. Frederick, and V. Jacobson, “RTP: a transport protocol for real-timeapplications,” RFC 3550, Internet Engineering Task Force, July 2003.

[67] J. Lennox and H. Schulzrinne, “Transporting user control information in SIP REGISTER pay-loads,” internet draft, Internet Engineering Task Force, Mar. 1999. Work in progress.

[68] M. A. Sasse, V. J. Hardman, I. Kouvelas, C. E. Perkins, O. Hodson, A. I. Watson, M. Handley, andJ. Crowcroft, “RAT (robust-audio tool),” 1995.

[69] A. Dutta, T. Zhang, S. Madhani, K. Taniuchi, K. Fujimoto, H. Schulzrinne, Y. Ohba, and Y. Kat-sube, “Secure universal mobility for wireless Internet,” in The Second ACM International Workshopon Wireless Mobile Applications and Services on WLAN Hotspots, Oct. 2004.

[70] A. Dutta, B. Kim, T. Zhang, S. Baba, K. Taniuchi, Y. Ohba, and H. Schulzrinne, “Experimentalanalysis of multi interface mobility management with SIP and MIP,” in IEEE Wirelesscom, IEEE,IEEE, June 2005.

[71] A. Dutta, T. Zhang, Y. Ohba, K. Taniuchi, and H. Schulzrinne, “MPA assisted proactive handoffscheme,” in ACM Mobiquitous, SIGMOBILE, ACM, 2005.34

[72] A. Dutta, S. Das, D. Famolari, Y. Ohba, K. Taniuchi, V. Fajardo, T. Kodama, and H. Schulzrinne,“Secured seamless convergence across heterogeneous access networks,” in World Telecommunica-tion Congress, (Budapest), IEEE, IEEE, May 2006.

[73] J.-O. Vatn and G. C. Maguire, “The effect of using co-located care-of addresses on macro han-dover latency,” in 14th Nordic Tele-traffic Seminar, (Technical University of Denmark, Lyngby,Denmark), Aug. 1998.

[74] A. Dutta, E. Van Den Berg, D. Famolari, V. Fajardo, Y. Ohba, K. Taniuchi, and H. Schulzrinne,“Dynamic buffering scheme for mobile handoff,” in PIMRC, (Helsinki), IEEE, IEEE, 2006.

[75] A. Dutta, D. Wong, J. Burns, R. Jain, K. Young, A. McAuley, and H. Schulzrinne, “Realization ofintegrated mobility management for ad-hoc networks,” in IEEE Milcom, (Annaheim, California),Oct. 2002.

[76] K. Wong, H.-Y. Wei, A. Dutta, K. Young, and H. Schulzrinne, “IP micro-mobility managementusing host-based routing,” in Wireless IP and building the Mobile Internet (S. Dixit and R. Prasad,eds.), Artech House, 2002.

[77] D. Wong, A. Dutta, J. Burns, K. Young, and H. Schulzrinne, “A multilayered mobility managementscheme for auto-configured wireless IP networks,” IEEE Wireless Magazine, Oct. 2003.

[78] D. Wong, A. Dutta, H. Schulzrinne, and K. Young, “Managing simultaneous mobility of IP hosts,”in IEEE International Military Communications Conference (MILCOM), (Boston, MA, USA),IEEE, IEEE, Oct. 2003.

[79] K. Wong, A. Dutta, H. Schulzrinne, and K. Young, “Simultaneous mobility: analytical framework,theorems, and solutions,” Wireless Communication and Mobile Computing, June 2006.

[80] D. A. Ferguson, “Measurement of mundane TV behaviors: Remote control device flipping fre-quency,” Journal of Broadcasting and Electronic Media, Jan. 1994.

[81] W. Fenner, “Internet group management protocol, version 2,” RFC 2236, Internet Engineering TaskForce, Nov. 1997.

[82] B. Williamson, Developing IP Multicast Networks: The definitive GUide to Designing and De-ploying CISCO IP Multicast Networks, vol. Volume 1. San Francisco: Cisco Press, Jan. 2000.

[83] M. Flament, F. Gessler, F. Lagergren, O. Queseth, R. Stridh, M. Unbehaun, J.-L. C. Wu, andJ. Zander, “An approach to 4th generation wireless infrastructures, scenarios and research issues,”in VTC, (Houston, Texas), May 1999.

[84] A. Dutta and H. Schulzrinne, “A streaming architecture for next generation Internet,” in ConferenceRecord of the International Conference on Communications (ICC), (Helsinki), p. 7, June 2001.

[85] M. Handley and V. Jacobson, “SDP: session description protocol,” RFC 2327, Internet EngineeringTask Force, Apr. 1998.

[86] M. Handley, C. E. Perkins, and E. Whelan, “Session announcement protocol,” RFC 2974, InternetEngineering Task Force, Oct. 2000.

[87] H. Schulzrinne, A. Rao, and R. Lanphier, “Real time streaming protocol (RTSP),” RFC 2326,Internet Engineering Task Force, Apr. 1998.

[88] S. Kaur, B. Madan, and S. Ganeshan, “Multicast support for mobile IP using modified IGMP,” inIEEE WCNC, pp. 948–952, Mar. 1999.

35

[89] A. Zimmermann, J. Freiheit, R. German, and G. Hommel, “Petri net modeling and performabil-ity evaluation with timeNET 3.0.,” Lecture Notes in Computer Science: Computer PerformanceEvaluation: Modeling Techniques and Tools, vol. 1786, pp. 188–202, 2000.

[90] R. K. Ed., “Fast handovers for mobile ipv6,” RFC 4068, Internet Engineering Task Force, July2005.

36