ieee transactions on vehicular

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008 1209

An Architecture for Seamless Mobility Support inIP-Based Next-Generation Wireless Networks

Christian Makaya, Student Member, IEEE, and Samuel Pierre, Senior Member, IEEE

Abstract—Recent technological innovations allow mobile de-vices to be equipped with multiple wireless interfaces. Moreover,the trend in fourth-generation or next-generation wireless net-works (4G/NGWNs) is the coexistence of diverse but complemen-tary architectures and wireless access technologies. In this context,an appropriate mobility management scheme, as well as the inte-gration and interworking of existing wireless systems, is crucial.Several proposals for solving these issues are available in the litera-ture. However, these proposals cannot guarantee seamless roamingand service continuity. This paper proposes a novel architecturecalled Integrated InterSystem Architecture (IISA), which is basedon the Third-Generation Partnership Project/Third-GenerationPartnership Project 2 requirements that enables the integrationand interworking of current wireless systems, and investigatesmobility management issues. An efficient handoff protocol basedon localized mobility management, access networks discovery,and fast handoff concepts called Handoff Protocol for IntegratedNetworks (HPINs) is proposed. It alleviates service disruption dur-ing handoff in IPv6-based heterogeneous wireless environments.Numerical results show that HPIN performs better in terms ofsignaling cost, handoff latency, handoff-blocking probability, andpacket loss compared to existing schemes.

Index Terms—Internet Protocol (IP)-based wireless networks,interworking architecture, mobility management, quality-of-service (QoS), seamless roaming, service continuity, verticalhandoff.

I. INTRODUCTION

FOURTH-GENERATION or next-generation wireless net-works (4G/NGWNs) are expected to exhibit heterogeneity

in terms of wireless access technologies, services, applicationrequirements, high usability, and improved capacity. With 4G/NGWNs, users will intensify demands for seamless roamingacross different wireless networks, support of various services(e.g., multimedia applications), and quality-of-service (QoS)guarantees. The strengths of third-generation (3G) cellularnetworks, such as UMTS and CDMA2000, consist of theirglobal coverage, whereas their weaknesses lie in bandwidthcapacity and operation costs. On the other hand, wireless localarea network (WLAN) technology, such as IEEE 802.11, offershigher bandwidth with low operation costs, although it covers arelatively short range. Moreover, technological advances in the

Manuscript received November 7, 2006; revised April 5, 2007 and June 14,2007. This work was supported in part by the NSERC/Ericsson IndustrialResearch Chair under Grant IRC252720-01. The review of this paper wascoordinated by Dr. Q. Zhang.

The authors are with the Mobile Computing and Networking Re-search Laboratory (LARIM), Department of Computer Engineering, EcolePolytechnique de Montreal, Montreal, QC H3C 3A7, Canada (e-mail:[email protected]; [email protected].).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TVT.2007.906366

Fig. 1. Overview of the 4G/NGWN architecture.

evolution of portable devices have made possible the supportof different radio access technologies (RATs). This has raisedmuch interest in the integration and interworking of 3G wirelessnetworks with WLANs due to the potential benefits of theircomplementarity. Evolution through this integration is one ofthe paths to next-generation wireless network (NGWN) designrather than investing efforts into developing new radio inter-faces and technologies [1]. Integrated networks will providethe benefits of both technologies to end users as well as toservice providers. The integration of wireless networks willnot be limited only to WLAN and 3G cellular networks butwill also be extended to other technologies, such as satellitenetworks, WiMAX, mobile ad hoc networks, and wirelesssensor networks.

Conceptually, a typical NGWN architecture can be viewed asmany overlapping wireless access domains, as shown in Fig. 1,and is called a wireless overlay network [2]. The main goal ofan NGWN is to allow subscribers to profit services anytimeand anywhere, which is known as always best connected [3].The heterogeneity in terms of RATs and network protocols inan NGWN asks for a common interconnection element. Sincethe Internet Protocol (IP) technology enables the support ofapplications in a cost-effective and scalable way, it is expectedto become the core backbone network of an NGWN [4]. Hence,current trends in communication network evolution are directedtoward the all-IP principle to hide heterogeneities and achieveconvergence of various networks.

Two major architectures (loose and tight coupling) for 3G/WLAN interworking based on existing 3G network architecturecomponents have been proposed in [5]. All scenarios presentedin [5] and [6] are not yet fulfilled, and those interworking archi-tectures have pros and cons. The integration and interworkingof heterogeneous wireless networks are widely documented inthe literature, and various models have been proposed. Both3G wireless network initiatives Third-Generation Partnership

0018-9545/$25.00 © 2008 IEEE

1210 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 2, MARCH 2008

Project (3GPP) and Third-Generation Partnership Project 2(3GPP2) have proposed a 3G/WLAN interworking architectureadapted to their respective systems. An evident way to achieveroaming among various networks is by using bilateral servicelevel agreements (SLAs). However, due to several reasons, thisapproach is not feasible. In fact, the increasing number ofwireless networks and service providers makes it impracticalfor network operators to have direct SLAs with all of the otheroperators. Moreover, network operators are reticent in makingtheir databases available to other operators.

Mobility management, with provision of seamless handoffand QoS guarantees, is one of the key issues to support globalroaming of mobile nodes (MNs) between various wirelesssystems in an efficient way. In an NGWN, mobility is not onlya physical concept but also a logical one. It is thus crucial toprovide seamless roaming and QoS guarantee support basedon intelligent and efficient mobility management schemes.To enable service continuity and QoS provision, a seamlesshandoff (i.e., minimal service disruption during handoff) is ofgreat importance. Seamless handoff means low packet loss,minimal handoff latency, low signaling traffic overhead, andlimited handoff failure. The handoff latency refers to the timeinterval during which an MN cannot send or receive any datatraffic during handoffs. It is composed of L2 (link layer) and L3(IP layer) handoff latency. The overall handoff latency may besufficiently long to cause packet loss, which is unacceptable forreal-time applications.

The QoS guarantee represents one of the major challeng-ing issues due to the heterogeneity of network architectures,network capacities, different high-layer protocols, and variousRATs. An exact mapping between all 3G wireless network QoSparameters and WLAN QoS parameters is highly difficult toperform and remains to be an open issue since these networksare totally different. The handoff process in an NGWN can besubdivided into three phases: 1) network discovery; 2) handoffdecision; and 3) handoff execution. The simplest way for anMN with multiple air interfaces to discover reachable wirelessnetworks is to keep all air interfaces on at all times. However,keeping an air interface continuously active consumes batterypower even while the mobile device is not sending/receivingpackets. It is thus critical to avoid keeping idle air interfaces per-petually on. Moreover, an MN must observe if the new networkis consistently better than the current one before performinghandoff to avoid the ping-pong effect.

In homogeneous wireless networks, handoff decisions aretypically driven by metrics that are strictly related to receivedsignal strength (RSS) quality and resource availability. How-ever, in an NGWN, RSSs from different networks do not sharethe same meaning since each network is composed of its spe-cific features; thus, they cannot be directly compared. Hence,handoff decisions with signal strength as the sole criterion maybe inefficient or impractical in an NGWN. More complex met-rics combining a higher number of parameters such as monetarycosts, bandwidth, power consumption, network conditions, anduser preferences must be defined [7].

This paper proposes a novel architecture called Integrated In-terSystem Architecture (IISA) based on 3GPP/3GPP2-WLANinterworking models to integrate existing wireless systems,

such as 3GPP/3GPP2, WLAN, and WiMAX, and hide their het-erogeneities. Furthermore, we propose a mobility managementscheme called the Handoff Protocol for Integrated Networks(HPINs), which provides QoS guarantees for real-time appli-cations in heterogeneous IPv6-based wireless environments.HPIN is a one-suite protocol that performs access networkdiscovery, fast handoff, and localized mobility management.HPIN allows the selection of the best available network atany given time, and it is designed for both heterogeneousand homogeneous wireless networks. In other words, the maincontributions of this paper are given as follows:

1) the design of an interworking architecture that permits theintegration of any type of wireless networks rather thanonly 3G cellular systems with WLAN or heterogeneous3G cellular systems;

2) the design of an efficient handoff management scheme,which enables the support of seamless handoff and ser-vice continuity for mobile users moving across variousnetworks;

3) the proposal of a new approach to speed up contexttransfers and binding updates (BUs) for mobile users;

4) the proposal of an analytical model to analyze the perfor-mance of the proposed mechanisms and architecture.

The remainder of this paper is organized as follows:Section II offers an overview of the basic concepts and relatedwork pertaining to interworking and mobility managementin heterogeneous IP-based wireless networks. Then, the pro-posed architecture (IISA) and the handoff management scheme(HPIN) are described in Sections III and IV, respectively. Theanalytical model is developed to assess the efficiency of HPINand IISA in Section V. Results from the performance evaluationare analyzed in Section VI before the concluding remarksdrawn in Section VII.

II. BACKGROUND AND RELATED WORK

Mobility management enables a system to locate roamingterminals to deliver data packets (i.e., location management)and maintain connections with them when moving into a newsubnet (i.e., handoff management). Handoff management is amajor component of mobility management since an MN cantrigger several handoffs during a session, as it will be thecase in an NGWN. Handoffs in an IP-based NGWN involvechanges of access points or base stations (AP/BSs) at thelink layer and possibly routing changes at the IP layer. Withthe coexistence of various wireless access technologies, twokinds of handoffs are possible in an NGWN: 1) horizontal and2) vertical handoffs. Horizontal or intrasystem handoffs occurwhen an MN is moving between AP/BSs of the same networktechnology. When AP/BSs belong to different networks (e.g.,IEEE 802.11 and UMTS), such a movement is called a verticalhandoff.

Two kinds of vertical handoffs can occur, depending on thetype of overlapping. In fact, roaming may happen between fullyoverlapping networks from low-tier (e.g., WLAN) to high-tier(e.g., 3G wireless network) networks and vice versa, or betweenpartially overlapping networks. In case of roaming under fullyoverlapping networks, vertical handoffs are usually asymmetric

MAKAYA AND PIERRE: ARCHITECTURE FOR SEAMLESS MOBILITY SUPPORT IN IP-BASED NGWNs 1211

and can focus on improving either the transmission rate orsession connectivity [2]. The characteristics of NGWN makethe implementation of vertical handoffs more challenging thanthat of horizontal handoffs. In fact, maintaining uninterruptedsessions while the physical interface is changing constitutes acomplex task. Several IPv6-based handoff protocols proposedin the literature to manage horizontal and vertical handoffsmay appear appropriate. However, they have advantages anddrawbacks, and have been separately proposed. Much work isstill required for further improvements in 4G/NGWNs.

A. IPv6-Based Mobility Schemes

Mobile IPv6 (MIPv6) [8] was proposed by the InternetEngineering Task Force (IETF) for mobility management atthe IP layer and allows MNs to remain reachable in spite oftheir movements within wireless IP environments. MNs arealways identified by their home address, regardless of theircurrent network point of attachment. While away from its homenetwork, an MN is associated with a care-of address (CoA),which provides information about its current location. Afteracquiring a CoA, an MN sends a binding update (BU) messageto the home agent (HA) to indicate its new address and also toall active correspondent nodes (CNs) to allow route optimiza-tion. However, MIPv6 has some well-known drawbacks suchas signaling traffic overhead, high packet loss rate, and handofflatency, thereby causing user-perceptible deteriorations of real-time traffic [9], [10].

These weaknesses led to the investigation of other solutionsto enhance MIPv6. Two main MIPv6 extensions proposed bythe IETF are Hierarchical MIPv6 (HMIPv6) [11] and FastHandovers for MIPv6 (FMIPv6) [12]. These protocols tacklemicromobility, whereas MIPv6 is used for macromobility.HMIPv6 locally handles handoff through a special node calledthe mobility anchor point (MAP). The MAP, acting as a localHA in the network visited by the MN, limits the amount ofMIPv6 signaling outside its domain and reduces delays that areassociated with the location updates. However, HMIPv6 cannotmeet the requirements for delay-sensitive traffic, such as voiceover IP, due to packet loss and handoff latency. FMIPv6 hasbeen proposed to minimize service disruption during handoffspertaining to MIPv6 operations, such as movement detection,BU, and addresses configuration. The link layer information(L2 trigger) is used either to predict or to rapidly respond tohandoff events.

Although FMIPv6 paves the way for the improvement ofMIPv6 performance in terms of handoff latency, it is still hin-dered by several problems, such as QoS support and scalability.In fact, FMIPv6 does not effectively reduce signaling overheadnor packet loss, which leads to unacceptable service disruption.In FMIPv6, the new access router (NAR) consumes storagespace to buffer forwarded packets by the previous access router(PAR) before delivering these packets to the MN. These for-warded packets lack QoS guarantee before the new QoS path isset up. Combining HMIPv6 and FMIPv6 motivates the designof Fast Handover for HMIPv6 (F-HMIPv6) [13] to increasenetwork bandwidth usage efficiency. However, F-HMIPv6 mayinherit the drawbacks of both FMIPv6 and HMIPv6, such as

synchronization issues and signaling overhead [9], [10]. Withthose IPv6-based mobility protocols, seamless mobility cannotbe guaranteed.

To achieve seamless mobility across various access technolo-gies and networks, an MN needs information about the wirelessnetwork to which it could attach. In addition, it is necessary totransfer information (context transfer) that is related to an MNfrom the current access router to the next one. To enable theseprocedures, the Candidate Access Router Discovery (CARD)Protocol [14] and the Context Transfer Protocol (CXTP) [15]have been proposed. They avoid using limited wireless re-sources and provide fast mobility and secure transfers. Theirkey objectives consist of reducing latency and packet losses,and avoiding the reinitiation of signaling to/from an MN fromthe beginning during a handoff. However, context transfer isnot always possible. For example, when an MN moves acrossdifferent administrative domains, the new network may requirethe MN to reauthenticate and perform signaling from the be-ginning rather than accepting the transferred context. With theCARD Protocol, acquiring L3 information of neighbor accessrouters (ARs) is based on L2 ID detection, which is possibleonly when the associated air interface is on. In addition, MNsmust periodically monitor the RSS from neighbor AP/ARs andconstruct a neighbor network information table. Moreover, en-tities exchanging contexts must authenticate each other, whichcould turn into a tedious procedure in 4G/NGWNs.

B. 3G/WLAN Interworking Models

Six 3G/WLAN interworking scenarios and their require-ments have been defined in [16] and [17] to provide a properbackground for interworking architecture design. With the par-ticular characteristics of WLAN and 3G wireless networks, twoscenarios present significant technical challenges: 1) servicecontinuity and 2) seamless roaming provision. To handle thesescenarios, two interworking architectures called loose and tightcoupling have been proposed by 3GPP [5]. With the tightcoupling approach, a WLAN appears as one of the 3G radioaccess networks to the 3G wireless core network. Althoughtight coupling allows easy control of QoS for time-sensitivetraffic, it includes several drawbacks, such as high costs andcomplexity levels. Moreover, with tight coupling, traffic from aWLAN flows into the 3G wireless core network and createscapacity problems. In fact, 3G wireless core network nodescannot accommodate the bulk of the data traffic from a WLAN.

On the other hand, with loose coupling, different networksare independently deployed, and data paths are completelyseparated between WLANs and 3G wireless networks. Hence,loose coupling enables several advantages, such as independenttraffic engineering, low costs, and low complexity levels. How-ever, loose coupling may not guarantee service continuity toother access networks during handoff as it suffers from longhandoff latency and packet loss. The choice of an optimalinterworking architecture is determined by a certain numberof factors. For example, if the wireless network is composedof a large number of WLAN and 3G network operators, theloosely coupled architecture would be the best choice. On theother hand, if the WLAN network is exclusively owned by a


3G wireless operator, the tightly coupled architecture mightbecome a more relevant option. However, loose coupling is themost advocate interworking scheme [18].

C. Handoff Management Schemes

In [18], an integrated architecture and a radio interface selec-tion scheme were proposed based on signal strength and radiointerface priorities. As previously mentioned, these param-eters are not appropriate for handoff decisions in an NGWN.Moreover, an MN must passively evaluate handoff conditions,even when the application in the current network is runningwell. This introduces unnecessary power consumption and net-work resource usage. HandOff Protocol for OVERlay networks(HOPOVER), a mobile IP-based approach, was proposed in[19] and handles both vertical and horizontal handoffs. Al-though HOPOVER enables a low signaling overhead, it re-quires APs to maintain an excessive quantity of informationabout MNs. An architecture for next-generation all-IP-basedwireless systems was proposed in [4]. Two new entities, i.e.,the network interworking agent and the interworking gateway(IG), are introduced to allow the integration of several wire-less networks while supporting MN roaming. However, thisproposed architecture provides no appropriate handoff decisionmechanism to take heterogeneity into account. The handoffdecision is based on the RSS criterion, which, as previouslymentioned, is not appropriate for an NGWN.

In [20], a mobility management scheme and an architecturewere proposed to support roaming across 3G heterogeneouswireless networks but not for IP-based wireless networks oran authentic NGWN. This proposed architecture is based ona boundary location register (BLR) and a border interworkingunit (BIU), which are placed at the border of two neighboringsystems. This approach is not scalable in the sense that oneBLR/BIU is needed for each pair of adjacent networks.Furthermore, directly connecting BIUs to the visitor locationregister (VLR) of each subsystem creates several drawbacks,such as the increase in cabling costs and signaling traffic due topaging procedure execution through the home location register(HLR) of both involved networks. The BLR/BIU model cannotmeet all of the main requirements (economics, scalability,transparency to heterogeneous access technologies, seamlessmobility support, and security) of any novel architecture. Anarchitecture and a mobility management scheme that improvethe performance of the BLR/BIU model were proposed in [21].Although an HLR and a VLR may be seen as the HA andforeign agent in IP-based wireless networks, respectively, theirfunctionalities significantly differ. We focus not only on mobil-ity and interworking issues between 3G wireless networks butalso on an authentic NGWN, i.e., IP-based wireless networks.

A policy-enabled handoff decision algorithm proposed in[22] is based on a cost function that considers several factors(e.g., bandwidth, power consumption, and monetary costs). Thecost function presented in [22] is very preliminary and cannothandle more sophisticated scenarios. In addition, cost functionevaluation could require high processing time and power. Tomaximize user QoS, McNair and Zhu [7] proposed handoffdecision algorithms for vertical handoff and identified metrics

that characterize an NGWN. However, the proposed cost func-tion could lead to singularity problems if connections becomefree of charge. Furthermore, the handoff instability problem andmobility management at the IP layer are ignored. Many othervertical handoff schemes are presented in [23]. The factorsconsidered in the previously cited papers are insufficient. Infact, information about authentication types, access networktypes, and roaming partners supported is not taken into account.Moreover, these studies do not provide a viable architectureframework for selection mechanisms nor business models forprospective deployment.

III. PROPOSED ARCHITECTURE FOR NGWN

A novel interworking architecture called IISA based on3GPP/3GPP2-WLAN interworking models is proposed andshown in Fig. 2. Instead of developing new infrastructures,IISA extends existing infrastructures to tackle integration andinterworking issues and provides mobile users with ubiquityor always best connected. The IISA considers all of theaforementioned requirements (i.e., scalability, transparency,economics, and security) for an IP-based NGWN. Rather thanadding an interworking entity between adjacent networks,as it is the case for some existing models presented in theliterature such as BLR/BIU, IISA only adds a single newnode called interworking decision engine (IDE), whereasother functionalities are implemented in the existing networkcomponents. Another main difference between our approachand the BLR/BIU architecture is the separation between thecontrol plane (signaling traffic) and the transport plane (datatraffic) in the IISA/HPIN proposal. In fact, only signalingtraffic, and not data packets, goes through the IDE. In theBLR/BIU architecture, data packets and signaling traffic transitthrough BLR/BIU, thus creating bottlenecks in the system.

To enable the support of IPv6-based mobility managementprotocols, some functional entities of 3G wireless networks areextended. The Serving General Packet Radio Service (GPRS)Support Node (SGSN) and Packet Control Function are en-hanced with the AR functionalities and are called the accessedge node (AEN). Similarly, the Gateway GPRS Support Nodeand Packet Data Serving Node are extended with MAP orHA and interworking functionalities (to enable message formatconversion, QoS requirements mapping, etc.) and are calledthe border edge node (BEN). The WLAN IG acts as a routepolicy element, ensuring message format conversion. Extendedfunctionalities can be integrated into existing network entitiesor separately implemented. We advocate the first scenario asit is easily deployed and managed. The interworking of dif-ferent access networks is required for an efficient integration.Mapping between the HLR or home subscriber server in 3Gwireless networks, and the authentication, authorization, andaccounting (AAA) server/proxy in a WLAN is required toexecute authentication and billing when users roam acrossboth technologies. A novel entity IDE is introduced to enableinterworking and handoff between various networks. Operatorsor service providers are required to have only one SLA with athird-party or IDE manager rather than establishing individualSLAs with all of the other operators.


Fig. 2. Integrated Intersystem Architecture (IISA).

Fig. 3. Interworking Decision Engine (IDE).

The IDE allows the reduction of signaling traffic and servicedisruption during handoff while handling AAA procedure andmobility management. To reduce the IDE’s load, the IDE isinvolved only in intersystem and/or interdomain handoff, andit manages only control signaling traffic: Data packet trafficbypasses the IDE. Furthermore, to enable the scalability of theIISA architecture, if the number of mobile users that requireintersystem and/or interdomain handoff increases, or if thenumber of heterogeneous wireless systems increases, the IDEcan be deployed within a hierarchical framework. For roamingusers with sessions in progress, the IDE allows the reduction ofassociation and authentication delays. Usage of the IDE couldbe considered as a value-added service that network operatorsoffer to their subscribers to allow roaming into other networks.

The logical components of the IDE are illustrated in Fig. 3.The Authentication Module (AuM) is used to authenticate usersmoving across different wireless networks, and it avoids therequired direct security agreements or associations between for-eign networks and the home network. The AuM stores informa-tion such as subscriber identity, user preferences, user profile,and terminal mobility patterns. The Accounting Module (AcM)enables billing between different wireless networks and storescharging information that is associated with the resource usage.

It acts as a common billing/charging system between variousnetwork operators. The AcM collects accounting informationreceived from the AAA server/proxy of the foreign networkper user based on its billing policy. If necessary, it convertsdetailed call records of the foreign network before forwardingsuch information to the AAA server of the home network forbilling purposes. The Cellular Intercarrier Billing ExchangeRoamer Record Protocol may be used for the exchange ofroaming billing information, for voice and data, among wirelesstelecommunication companies through the IDE.

Usually, different administrative domains have different QoSpolicies for resource allocation. Thus, when an MN moves fromone administrative domain to another, QoS renegotiation maybe required. Such renegotiation will be based on SLAs betweenboth domains. The Resource Management Module (RmM) en-ables QoS mapping and renegotiation. Furthermore, the RmMallows fast transfers of user profile and QoS requirements/parameters between two administrative domains during hand-off. The QoS mapping and the mechanism by which the IDEallocates resources to an MN and decides to admit a newrequest are outside the scope of this paper. However, we assumethat the IDE is endowed with intelligence and can performthe following operations: 1) translation of signaling messageformats between different networks; 2) conversion of highertransmission rate to lower rate; and 3) translation of QoSparameters and information, etc. The SLRA Module storesinformation about service providers or network operators thathave SLAs and roaming agreements with the IDE manager. TheHandover DecisionModule is used in deciding if an intersystemor interdomain handoff should be granted or not. In other words,it enables roaming and handoff support for MNs.

IV. PROPOSED HANDOFF PROTOCOL

Since mobility in NGWN is either logical or physical, userprofile and preferences seem to be important when performing


vertical handoff. As previously mentioned, handoff decisionsbased on the RSS level are not appropriate in 4G/NGWNs. In[24], a handoff decision function is proposed for handoff deci-sions, which take into account several parameters such as mon-etary cost, bandwidth, session priority, power consumption, andnetwork conditions to enable efficient decisions and systemdiscovery. In the following sections, we propose a handoffmanagement protocol that supports both vertical and horizontalhandoffs in IPv6-based heterogeneous mobile environments.Under IISA, intra-MAP/BEN and inter-MAP/BEN roamingmay result in either intrasystem or intersystem handoffs. Hence,HPIN is proposed for any of these types of handoff scenarios.

A. Authentication of MNs

To avoid additional signaling overhead due to the executionof the AAA procedure each time an MN performs handoffand requests registration, we propose a token-based approach.While roaming within the MAP/BEN domain of access net-works having agreements with the IDE, an MN presents a tokenthat it obtained from the IDE (after its first successful registra-tion in the visiting network) to the MAP/BEN or AR/AEN. Thetoken includes security association parameters for secure tunnelsetups between the MN and AR/AENs. This yields a lowerregistration latency than performing authentication and autho-rization check with the AAA home server. If the MAP/BENor AR/AEN successfully verifies the token, it initiates an au-thorization process. The HA functionalities that are related tothe MN authentication, distributing keying materials, sessionkeys, security association context, and mobility managementare delegated to the IDE while the MN roams in foreignnetworks. Subsequent authentications are handled either by theMAP/BEN and the AAA local server/proxy (AAAL) or by theIDE for intrasystem or intersystem movements.

B. Handoff Preparation With HPIN

With the assumption that mobile devices will become in-creasingly powerful, intelligent, and sensitive on changes of thelink layer, we adopt a network-assisted and mobile-controlledhandoff strategy. The proposed handoff scheme combinesmobile-monitored and network-probed information to providereliable handoff control. Prior to handoff, an MN can obtain theinformation of candidate wireless networks to which it is likelyto handoff and uses this information to optimize the handoffperformance. On the other hand, if mobile device capabilitiesare limited, handoff decisions are taken by mobility agents onthe network side (e.g., BEN or IDE).

The MN decides whether to send the CARD Request mes-sage to the MAP/BEN according to the generation of antic-ipated triggers (ATs). For example, high bit error rate, linkgoing down, weak signal strength, security risks, monetarycost, and geographical location can be used as ATs. Upongenerating ATs, the MN sends a CARD Request message con-taining user preferences, application-required QoS capabilitiesto its serving MAP/BEN. With this message, the MN requestsinformation on the neighbor networks of its serving networkfrom the IDE through the current MAP/BEN. With information

exchanged between the MAP/BEN and candidate AR/AENs(CAR/AENs) by using the Router Information eXchange (RIXRequest/Reply) messages, the MAP/BEN maintains a globalview (i.e., the load status of AR/AENs, and connection state ofany MN in its domain, and movement patterns of all its servingMNs) of its domain and can learn both link layer (L2) and IPlayer (L3) information in an access network. Note that, if theCARD Request message was not sent in time, for example,after generating the AT and before generating the L2 trigger,HPIN will turn into HMIPv6. However, if the CARD Requestmessage was not sent after generating the L2 trigger, the HPINturns into either an FMIPv6 or an F-HMIPv6.

L2 information may include the specific wireless accesstechnology and the system parameters (e.g., channel frequencyand number). On the other hand, L3 information may includethe AR/AEN global address, the prefix of the address adver-tised in wireless networks, and the current QoS status andparameters. The QoS status includes bandwidth availabilityand signal strength, whereas the QoS parameters may includeinformation such as supported data rate, video coding rate,and maximal delays. L2 and L3 information is then forwardedto the IDE and allows it to maintain a global view of allMAP/BEN domains having SLAs with the IDE manager. Toallow seamless service continuity, the requirements specifiedin the CARD Request message need to be consistently set upwith the QoS negotiated in the previous subnet/subsystem. TheQoS consistency is a highly challenging and crucial issue forreal-time applications. This consistency is handled by the IDE,which allows QoS mapping between various networks. Withthis information, prefiltering is performed by the MAP/BEN,based on the MN’s preferences, application-required capabil-ities, network availability, and the CAR/AEN list obtained. Ifthe MAP/BEN lacks user profile information, it requests suchinformation from the IDE rather than from the MN’s HA, whichis usually far away from the current MAP/BEN domain.

The MAP/BEN responds to the MN through a CARD Replymessage that contains the list of CAR/AEN. Upon receivingthe CARD Reply message, the MN configures new on-linkCoAs (NLCoAs) based on the stateless IPv6 address autocon-figuration mode [25]. The MN can then start the handoff atany time. CARD Request and Reply message exchange nolonger delay the handoff procedure, as it is performed whilethe MN uses the previous on-link CoA (PLCoA). Wheneverthe L2 trigger is generated, using the information providedby the CARD Reply message, the MN can select which airinterface to turn on for access network discovery and handoffpreparation. The L2 scanning process will be performed basedon the information provided in the CAR/AEN list rather thanthe scanning of all frequencies or channels. Then, the systemdiscovery and L2 scanning process can be accelerated. Thisselective interface activation enables better tradeoff between thesystem discovery time and power consumption efficiency com-pared to the always-on approach as used in most IPv6-basedmobility management protocols. The MN will then computethe handoff decision function [24] for each reachable networkcontained in the CAR/AEN list to determine whether a moresuitable network with a superior QoS level is available andselect it as a target network.


C. Handoff Execution With HPIN

After the previous step, the MN sends a fast BU (FBU)message to the serving MAP/BEN to notify the MAP/BENthat it is moving into a new subnet/subsystem. Upon receivingthe FBU, the MAP/BEN starts a fast handoff procedure bysending a handoff initiate (HI) message to NAR/AEN, whichincludes a request to verify the preconfigured NLCoA and toestablish a bidirectional tunnel between the MAP/BEN andNAR/AEN to prevent routing failure during handoff. In re-sponse to the HI message, the NAR/AEN performs a duplicateaddress detection (DAD) procedure before responding with ahandoff acknowledgment (HAck) message. After receiving theHAck message, the MAP/BEN sends the result to the MN byusing a fast BU acknowledgment (FBAck) message. Since theexact time when the MN will perform the link layer handoff isunpredictable, the FBAck message is sent to both the previousand new links. This ensures that the MN receives the FBAckmessage via either the PAR/AEN or NAR/AEN, confirming thesuccessful binding. Moreover, the MAP/BEN binds the PLCoAand NLCoA, and it tunnels any packets addressed to PLCoAtoward the NLCoA in the NAR/AEN’s subnet. The NAR/AENbuffers these forwarded packets until the MN becomes attachedto the NAR/AEN link.

The MN announces its presence on the new link by sendingrouter solicitation with the fast neighbor advertisement (FNA)option to the NAR/AEN. The FNA message is also used toconfirm the usage of NLCoA when the MN has not receivedthe FBAck message through the previous link. Optionally, theNAR/AEN responds to the FNA message with a neighbor ad-vertisement acknowledgment message to notify the MN to useanother NLCoA, contained in FBAck rather than its prospectiveNLCoA, if there is address collision. Then, the NAR/AEN willstart delivering buffered packets to the MN, with FBAck mostprobably as the first packet on the new link. The bidirectionaltunnel remains active until the MN completes the BU proce-dure. Note that, if an FBU message was not sent before the L2handoff, then an MN sends it piggybacked in the FNA message(FNA[FBU]) over the new link. When the NAR/AEN receivesthe FNA[FBU] message, it processes the FNA message part,extracts the FBU message part, and forwards it to the servingMAP/BEN. When the serving MAP/BEN receives the FBU,it responds by sending the FBAck message to NAR/AEN. Atthis time, the MAP/BEN can start tunneling toward NLCoAthe incoming and in-flight packets addressed to PLCoA. Thisprocedure refers to the reactive mode of HPIN, whereas thepredictive mode is previously explained (i.e., the MN sends theFBU through the PAR/AEN’s link and the FBAck is receivedbefore the L2 handoff). The reactive mode can either be inten-tionally carried or serve as a fallback solution when a predictivemode could not be successfully completed, for example, if theL2 handoff was completed before the FBAck message wasreceived by the MN.

In case of inter-MAP/BEN roaming, the bidirectional tunnelis established between the previous MAP/BEN (MAP1/BEN inFig. 5) and the NAR/AEN through the candidate MAP/BEN(MAP2/BEN in Fig. 5). Hence, the HI message is piggybackedin the handoff request (HOReq) message and sent to the candi-date MAP/BEN, which processes the HOReq message part, ex-

Fig. 4. Signaling message sequence with the HPIN for intra-MAP/BENroaming.

tracts the HI message, and forwards it to the target NAR/AEN.In response to the HI message, the NAR/AEN performs theDAD procedure before sending the HAck message. When thecandidate MAP/BEN receives the HAck message, it includesthis message in the handoff reply (HORep) message beforeforwarding it to the current MAP/BEN. After receiving HAck,the current MAP/BEN sends the result to the MN by usingFBAck on both links (previous and new) and establishes bind-ing between the previous and the new regional CoA (PRCoAand NRCoA), and tunnels any packets (buffered and incoming)addressed to PRCoA toward NRCoA. Message flow diagramsfor both the intrasystem or intersystem handoff during intra-MAP/BEN or inter-MAP/BEN roaming are illustrated in Figs. 4and 5, respectively.

D. Context Transfer and BU

Note that the HI message triggers the request of contexttransfer rather than the use of the Context Transfer ActivateRequest message, as it is the case in the CXTP [15]. Whenthe MAP/BEN receives an FBU message, it transmits a Con-text Transfer Data (CTD) message, piggybacked in HI, to theNAR/AEN containing feature contexts. Examples of featurescontained in the CTD message are QoS context information,header compression, security, and AAA parameters. This papermainly focuses on QoS context information. The routers extractthis QoS context information, and according to the contextreceived, the intermediate router reserves corresponding re-sources and updates the path information. If the MAP/BEN hasno context pertaining to the concerned MN, the new MAP/BENsends a Context Transfer Request (CTReq) message to theIDE to obtain session management parameters for this MNand to establish traffic bearers on the new path. In responseto a CTReq message, the IDE transmits a CTD message thatincludes the MN’s previous IP address [i.e., regional CoA(RCoA)] and feature contexts. When the MAP/BEN receivesa CTD message, it installs the contexts as received from theIDE. The MAP/BEN includes the CTD message within the HImessage and forwards it to the NAR/AEN.


Fig. 5. Signaling message sequence with the HPIN for inter-MAP/BEN roaming.

TABLE INOTATION

When the NAR/AEN receives the CTD message, it mayoptionally generate a CTD Reply (CTDR) message to report theprocessing status of the received contexts and piggybacks thismessage in HAck. The NAR/AEN will send a HAck message tothe MAP/BEN only after relocating traffic bearers and resourcereservation (the resource reservation procedure is out of thescope of this paper) toward the new path to indicate that handoffmay be conducted and packet forwarding may start. Hence,unlike FMIPv6 and F-HMIPv6, where forwarded packets haveno QoS guarantee before the new QoS path is set up, theHPIN solves this issue. The BU procedure is performed by theNAR/AEN on behalf of the MNs. In fact, an AR/AEN actsas a proxy, i.e., it copies a BU list of an MN in its cacheand manages this list (e.g., lifetime entries) in the same wayas the original is managed by the MN. The AR/AEN cachecopy must be periodically updated according to the originalBU list of the MN. The BU list contains information about a

used home address and CoA [on-link CoA (LCoA) and RCoA],IPv6 address of CNs, sequence number, lifetimes, and stateof retransmissions. When the BU list lifetimes cached in theAR/AEN are about to expire, the AR/AEN may send a BU listrenewal request to the MN. The BU list renewal is performedin the same way as a classical BU refresh [8]. By piggybackingthe BU list in an FNA message, separate out-of-band messagesfrom MN to NAR/AEN are avoided, thus reducing signalingtraffic overhead.

V. ANALYTICAL MODEL FOR HPIN

In IP-based wireless networks, QoS may be defined bypacket loss, handoff latency, handoff-blocking probability, andsignaling overhead. Analyses of these metrics are very useful inevaluating the performance of mobility management protocols.The notation used in this paper is given in Table I.


Let χT be the random variable for the time between the L2trigger generation and the link down (i.e., pending L2 handoff),and let fT (u, σ) be the probability density function for the suc-cessful completion of signaling, where σ > 0 is a success rateparameter. The probability Ps of anticipated handoff signalingsuccess for a particular observed valued tT is expressed by

Ps = Pr(χT > tT ) =

∞∫tT

fT (u, σ)du. (1)

Deriving an expression for Ps is difficult since it depends on theexact form of fT (u, σ), which is usually unknown. For the sakeof simplicity, we assume that χT is exponentially distributed.

A. User Mobility and Traffic Models

User mobility and traffic models are crucial for efficientsystem design and performance evaluation. Usually, an MNmobility is modeled by the cell residence time, and variousrandom variable types are used for this purpose [26]. Twocommonly used mobility models in wireless networks arerandom-walk and fluid-flow models [27]. Evaluating the timespan that an MN will stay within the subnet is usually based ontwo distributions: 1) exponential and 2) Gamma. The Gammadistribution is very realistic for mobility models as it considerschanges in the MN speed and direction. Note that exponentialdistribution is a particular case of Gamma distribution.

On the other hand, although the incoming calls or sessionsin an NGWN follow a Poisson process (i.e., the interarrivaltime are exponentially distributed), the interservice time is notnecessarily exponentially distributed [26]. Other distributionmodels, such as hyper-Erlang and Pareto, have been proposed.Furthermore, the self-similar nature of data traffic has beennoticed. However, performance evaluations reported in theliterature show that the exponential model can be appropriatefor cost analyses. In fact, the exponential model providesan acceptable tradeoff between complexity and accuracy.Hence, most cost analyses adopt exponential assumption [26].We consider a traffic model that is composed of two levels:1) session and 2) packets. The session duration follows an ex-ponential distribution with intersession rate λs, whereas packetgeneration and arrival rate follow a Poisson process.

Let µc and µd be the border crossing rate for an MN outof a subnet (i.e., AR/AEN domain) and a MAP/BEN domain,respectively. When an MN crosses a MAP/BEN domain bor-der, it also crosses an AR/AEN border. Then, let µl be theborder crossing rate for which an MN still stays in the sameMAP/BEN domain, i.e., µl = µc − µd. Under the fluid-flowmobility model, let v represent the average velocity of an MN,ρ be the user density, and Lc express the perimeter of a subnet.The subnet crossing rate can be computed by µc = ρvLc/π.If we assume that all subnets have a circular shape and formtogether a contiguous area and that each MAP/BEN domain iscomposed of M equally subnets, we obtain µd = µc/

√M [28].

Modeling of the probability distribution of the number ofboundary crossings during a session lifetime plays a significantrole in cellular network cost analyses. The same will apply to

IP-based wireless networks. For the sake of simplicity and toeasily derive analytical expressions, the exponential distributionwill be used. The roaming probability depends on an MN’smovement pattern in its original network but not in its desti-nation network. Hence, the probability that there is at least onelocal (resp. global) BU between two consecutive sessions of anMN Pc (resp. Pd) is expressed by

Pc = Pr(ts > tc) =

∞∫0

Pr(ts > u)fc(u)du =µc

µc + λs

Pd = Pr(ts > td) =

∞∫0

Pr(ts > u)fd(u)du =µd

µd + λs.

(2)

The probabilities that an MN experiences k subnet boundarycrossings and m access network boundary crossings duringthe lifetime of its session correspond to the probability massfunction of random variables Nc and Nd, respectively, and areexpressed by [29]

Pr(Nc = k) = P kc (1 − Pc)

Pr(Nd = m) = Pmd (1 − Pd). (3)

Then, the average number of location BUs during an in-tersession time interval under subnet crossings E(Nc) andMAP/BEN domain crossings E(Nd) are shown by

E(Nc) =∞∑

k=0

kPr(Nc = k)=∞∑

k=0

kP kc (1−Pc) =

µc

λs

E(Nd) =∞∑

m=0

mPr(Nd = m)=∞∑

m=0

mPmd (1−Pd)=

µd

λs.

(4)

With the same assumption on time variables, we can obtainthe expression of E(Nl), i.e., the average number of subnets(AR/AENs) that an MN crosses and that still stay within agiven MAP/BEN domain during an intersession time interval,as follows: E(Nl) = µl/λs.

B. Total Signaling Cost

Performance analyses of wireless networks must consider atotal signaling cost induced by a mobility management scheme.As for wireless cellular networks, signaling traffic overheadcost must be computed for an NGWN or IP-based mobileenvironments. An NGWN supports two kinds of locations orBUs. One occurs from an MN’s subnet boundary crossing, andthe other occurs when the binding lifetime is about to expire.Moreover, data packet delivery induces the use of networkresources, thus generating an additional cost. Hence, the totalsignaling cost CT could be considered as the sum of BUsignaling cost CBU and packet delivery cost CPD, and givenby CT = CBU + CPD.


TABLE IIEXPRESSION OF SIGNALING COSTS

Fig. 6. Timing diagram of the HPIN for intra-MAP/BEN roaming.

1) BU Signaling Cost: Depending on the movement type,two kinds of BU can be performed: 1) local and 2) global. Theglobal BU occurs when an MN moves out of its MAP/BENdomain. In this case, the MN registers its new RCoA to HAand to active CNs. On the other hand, if the MN changes itscurrent address (LCoA) within a MAP/BEN domain, it onlyneeds to register this new LCoA to the MAP/BEN. Hence,the average BU signaling cost during intersession time inter-vals heavily depends on the computed number of BUs and isgiven by

CBU = E(Nl)Cl + E(Nd)Cg. (5)

To perform signaling overhead analyses, a performance fac-tor called session-to-mobility ratio (SMR) is introduced. It issimilar to the call-to-mobility ratio defined in wireless cellularnetworks [30]. The SMR represents the relative ratio of thesession arrival rate over the user mobility rate: SMR = λs/µc.The BU signaling cost CBU is then given by

CBU =1λs

(µdC

g + µlCl)

=1

SMR√

M

[Cg + (

√M − 1)Cl

]. (6)

Anticipated trigger and link layer information (L2 trigger)is used to either predict or rapidly respond to handoff events.Hence, the HPIN signaling cost depends on the probability thathandoff anticipation is relevant or not. The critical phase of theHPIN starts when the L2 trigger is generated and indicates theimminence of the handoff. We assume that, if an MN receivesan FBAck message from the MAP/BEN, it will definitely startthe L3 handoff to NAR/AEN without exceptions. Hence, ifthere is no real handoff after the L2 trigger, all messages

exchanged for handoff anticipation may be unnecessary. Thus,global and local BU signaling costs for the HPIN are expressedas follows:

Cg =PsSgs + (1 − Ps)

(Sg

f + Sgr

)+ Cru

Cl =PsSls + (1 − Ps)

(Sl

f + Slr

)+ Cmu (7)

where Cru represents the BU cost at the IDE or at HA/CNs,Cmu depicts the BU cost at MAP/BEN, Sg

s (resp. Sls) de-

notes the global (resp. local) signaling cost for successfullyanticipated handoff, Sg

f (resp. Slf ) denotes the global (resp.

local) signaling cost for the control messages bore if no realL3 handoff occurs, and Sg

r (resp. Slr) indicates the global

(resp. local) signaling cost for the HPIN reactive mode. Theexpressions of those partial signaling costs are given in Table II.2) Packet Delivery Cost: Similar to the investigation re-

ported in [31], the handoff latency is subdivided into threecomponents: 1) link switching or L2 handoff latency tL2;2) IP connectivity latency tIP due to movement detectionand address configuration; and 3) location update latency tU .The IP connectivity latency reflects how quickly an MN cansend IP packets after L2 handoff, whereas the location updatelatency is the delay required to forward IP packets to the MN’snew IP address. On the other hand, the time period from thestarting point of the L2 handoff to when an MN receives IPpackets for the first time after link switching refers to packetreception latency tP or data latency. Moreover, the followingdelay components are defined: BU latency tBU and delay fromthe completion of BU and reception of the first packet by anMN through the new IP address tNR.

When two endpoints have an ongoing session, a packetdelivery cost is incurred. The packet delivery cost is composedof packet transmission and packet processing costs. By usingthe handoff timing diagram illustrated in Fig. 6, the packet


Fig. 7. Timing diagram of the HPIN for inter-MAP/BEN roaming.

delivery cost could be defined as the linear combination ofpacket tunneling cost Ctun and packet loss cost Closs. Let αand β be the weighting factors that emphasize the tunneling anddropping effects. The packet delivery cost CPD is computed asfollows:

CPD = αCtun + βCloss. (8)

To avoid packet loss, the HPIN enables the MAP/BEN toforward packets to NAR/AEN by using a tunnel established be-tween them, and the NAR/AEN buffers all forwarded packets.The HPIN timing diagram for intra-MAP/BEN movement isshown in Fig. 6.

In IP networks, the signaling cost is proportional to thedistance in hops between the source and destination nodes. Fur-thermore, the transmission cost in a wireless link is generallylarger than the transmission cost in a wired link [30]. Let sc

and sd be the average size of control packets and data packets,respectively, and η = sd/sc. The cost of transferring a datapacket is η greater than the cost of transferring a control packet.Let λp be the packet arrival rate expressed in packets per time.The packet tunneling cost for the HPIN predictive mode can beexpressed as follows:

Cp,ltun = λpC

s,lcm

(tL2 + tPIP + tLU

)(9)

where Cs,lcm = η(CCN,BEN + CBEN,NAR + CNAR,MN) is the

cost of transferring data packets from the CN to an MN bytransiting to MAP/BEN, tLU denotes the location update latencyfor intra-MAP/BEN movement (tLU = tLBU + tLNR), and tPIPdepicts the IP connectivity latency, excluding the IP addressconfiguration, DAD procedure, and movement detection. Infact, these operations are conducted before an MN leaves thePAR/AEN’s link.

In fast handoff schemes, packet losses are due to eitherL2 handoffs or an MN moving to another subnet before aforwarding tunnel has been established. The latter case refersto wrong temporal and spatial predictions. Packet loss dueto L2 handoff delay is inevitable without efficient bufferingmechanisms [31]. Let tLT be the time required to establisha tunnel between the MAP/BEN and the NAR/AEN. Usually,tT is greater than tLT ; thus, packets received during handoffare forwarded to the NAR/AEN by the MAP/BEN using thealready established tunnel. However, if the MN moves veryfast, tT may be inferior to tLT . Then, packets arriving to the

MAP/BEN during time period tLT−tT may be lost since thetunnel is not yet established. In other words, for the anticipatedsignaling to succeed, the following time constraint must beobserved: tLT ≤ tT . Hence, the cost associated with the packetloss can be expressed as follows:

Cp,lloss = λpC

f,lcm max(tLT − tT , 0) (10)

where Cf,lcm = η(CCN,BEN + CBEN,PAR + CPAR,MN) is the

cost of transferring data packets from the CN to the MN bytransiting to MAP/BEN when the handoff fails or if the BU isnot yet performed at the MAP/BEN.

Due to the wrong spatial predictions of NAR/AEN, or ifan FBAck message was not received through the previouslink, packets forwarded to a mispredicted NAR/AEN by theMAP/BEN may be lost. The process of forwarding packets tothe wrong NAR/AEN is stopped when the FBU message sentthrough the NAR/AEN’s link is received at the MAP/BEN.Moreover, if an MN’s movement within the subnet overlap-ping area is longer than the tunnel establishment delay, theHPIN turns into its reactive mode. Since the packet forwardingprocess is not supported in the reactive mode, the packettunneling cost is equal to zero (Cr,l

tun = 0), whereas the HPINpacket loss cost can be expressed as follows:

Cr,lloss = λpC

f,lcm

(tL2 + tRIP + tLU

)(11)

where tRIP is the IP connectivity latency of the reactive modefor an intra-MAP/BEN movement. The average packet deliverycost of the HPIN scheme is then given by

Ca,lPD = PsC

p,lPD + (1 − Ps)C

r,lPD (12)

where Cp,lPD and Cr,l

PD are the packet delivery costs for the HPINpredictive and reactive modes and are computed by (8).

The timing diagram of the HPIN for inter-MAP/BEN roam-ing is illustrated in Fig. 7. With similar reasoning as for intra-MAP/BEN, the packet tunneling cost Ctun and packet loss costCloss for inter-MAP/BEN roaming with the HPIN are expressedas follows:

Cp,gtun = λpC

s,gcm

(tL2 + tPIP + tU

)Cp,g

loss = λpCf,gcm max(tGT − tT , 0)

Cr,gloss = λpC

f,gcm

(tL2 + tRG

IP + tU)

(13)


where, for inter-MAP/BEN roaming, tU = tBU + tRR + tNR;tRR is the delay to complete the return routability procedure;tRGIP is the IP connectivity latency for the reactive mode; and

tGT is the time required to establish a tunnel between theprevious MAP/BEN and the NAR/AEN, i.e.,

Cs,gcm = η[CCN,pBEN + CpBEN,nBEN]

+ η[CnBEN,NAR + CNAR,MN]

Cf,gcm = η(CCN,pBEN + CpBEN,PAR + CPAR,MN).

C. Handoff Latency and Packet Loss

The following parameters are defined to compute handofflatency and packet loss: tL2 indicates the L2 handoff latency,and tX,Y specifies one-way transmission delay between nodesX and Y for a message of size s. If one of the endpoints is anMN, tX,Y is computed as follows:

tX,Y (s) =1 − q

1 + q

(s

Bwl+ Lwl

)

+ (dX,Y − 1)(

s

Bw+ Lw + q

)(14)

where q is the wireless link failure probability, q is the averagequeuing delay at each router in the Internet [32], Bwl (resp. Bw)denotes the wireless (resp. wired) link bandwidth, and Lwl

(resp. Lw) expresses wireless (resp. wired) link delay.Let ∆ns be the time elapsed between the reception of the

FBAck on the previous link and the beginning of the L2 handoffwhen L2 and L3 handoff operations are not well synchronized.Moreover, let ∆lr be the time between the last packet receptionthrough the previous link and the beginning of the L2 handoffwhen FBAck is received on the new link. Note that ∆lr and ∆ns

may be equal to zero. For HPIN, the handoff latency dependson the available information and on which link fast handoffmessages are exchanged. If information about NAR/AEN andan impending handoff are available, and if the FBAck messageis received through the previous link, the handoff latency isexpressed as follows:

OlHPIN = ∆ns + tL2 + 2tMN,NAR. (15)

However, if an FBAck message is not received through theprevious link, it will be received through the new link. In thiscase, the handoff latency for the HPIN is expressed as follows:

N lHPIN = ∆lr + tL2 + 2tMN,NAR + 3tNAR,BEN. (16)

The average handoff latency with the HPIN for intra-MAP/BEN roaming is given as follows:

DlHPIN = PsO

lHPIN + (1 − Ps)N l

HPIN. (17)

For the inter-MAP/BEN roaming case, when the FBAckmessage is received through the previous link, the handoff la-tency associated to HPIN is identical to intra-MAP/BEN roam-ing: Og

HPIN = OlHPIN. In fact, the handoff procedure depends

only on intra-MAP/BEN communication delay since the inter-MAP/BEN signaling is completed before the L2 handoff. Onthe other hand, when the FBAck message is received throughthe new link for inter-MAP/BEN movement, we assume thatappropriate information about the NAR/AEN is already avail-able and that NLCoA is already configured. Hence, the handofflatency with HPIN for inter-MAP/BEN roaming is given by

NgHPIN = ∆lr + tL2 + 2tMN,NAR

+ 3[tNAR,nBEN + tnBEN,pBEN]. (18)

The average HPIN handoff latency for inter-MAP/BEN roam-ing is similarly computed as in (17). With the HPIN, in theory,no packets are lost, unless buffers overflow at NAR/AEN orMAP/BEN. However, without efficient buffer management,forwarded packets can be lost during handoff latency. In fact,the number of packets lost is proportional to the handoff latency.

D. Handoff-Blocking Probability

The handoff-blocking probability is used to express thelikelihood that a session/call connection will be prematurelyterminated due to unsuccessful handoff during a session life-time. Subscribers are more sensitive to session blocking duringhandoff than at the moment the call was initiated. Hence, theminimization of the handoff-blocking probability is crucial formobility management schemes. The handoff blocking can becaused by many factors, including handoff latency, signal-to-noise ratio deterioration, unavailable channel, and session re-jection by the target network. However, this analysis considersonly latency as a handoff-blocking factor.

When an MN moves from one subnet to another, if the subnetresidence time is less than the total handoff time, packets arelost, and service is forcefully terminated due to loss of link orchannel. Let TS be the random variable defining the signalingdelay due to the handoff and T̃S be the mean value of the totalhandoff latency. If we assume that TS is exponentially dis-tributed with cumulative density function FT (t), the handoff-blocking probability is given by

PB =Pr(TS > tc)=

∞∫0

[1 − FT (u)] fc(u)du=µcT̃S

1 + µcT̃S

.

(19)

E. Processing Load of the IDE

Wireless overlay networks are subdivided into low-tier (e.g.,WLAN) and high-tier (e.g., 3G wireless network) networks[2]. Roaming between low-tier and high-tier networks refers tovertical or intersystem handoff. To analyze the load incurredat the IDE, we assume that high-tier networks fully overlaplow-tier networks and that users are uniformly distributed. LetNh and Nl be the number of high-tier and low-tier networksin the service or coverage area (e.g., one city), respectively.User density is denoted by ρh and ρl in high-tier and low-tiernetworks.

Recall that, with MIPv6, each subnet crossing results in a BUto the HA. Moreover, during the refresh time period, each MN


TABLE IIIPERFORMANCE ANALYSIS PARAMETERS

sends out a refresh request to the HA. Thus, the processing loadat the HA with MIPv6 scheme is given as follows:

LHA = PBU[NlρlvlLl + NsρhvhLs]

π

+PBR[νlρlAlNl + νhρhAhNh]

THA(20)

where Ns is the total number of subnets in a high-tier network,Nh ≤ Ns, νl (resp. νh) stands for the proportion of subscribersin a low-tier (resp. high-tier) network away from their homenetwork, PBU is the processing time for an update registrationmessage, and PBR depicts the processing time for the bindingrefresh message. THA and TIDE denote the binding lifetimeperiod at the HA and the IDE, respectively, whereas Al and Ah

indicate the coverage area of low-tier and high-tier networks,respectively. On the other hand, vl and vh are the average speedsof an MN in low-tier and high-tier networks, respectively,whereas Ll is the perimeter of the low-tier network and Ls theperimeter of a subnet in the high-tier network.

In the HPIN, binding refresh and BU are locally performedat the MAP/BEN and not to the IDE as long as an MNmoves within the MAP/BEN domain or performs intrasystemhandoffs. However, during the refresh time period TIDE, theMAP/BEN sends one RIX (Request or Reply) message to theIDE for a given number of MNs. We denote the number of theseMNs for low-tier networks as εl and that for high-tier networksas εh. Therefore, when intersystem and/or interdomain handoffoccurs, path updates are required. Thus, the IDE processingload is expressed as follows:

LIDE = PPU[NlρlvlLl + NhρhvhLs]

π

+PPR

⌈νlρlAlNl

εl

⌉ ⌈νhρhAhNh

εh

⌉TIDE

(21)

where PPU stands for the processing time for path updates,and PPR is the processing time for the path refresh message.Comparing (20) and (21) clearly shows that LIDE ≤ LHA.

Fig. 8. Network topology used for analysis.

On the other hand, assume that there are O operators in theservice area. The number of bilateral SLAs required to realizeroaming among all the networks deployed with a traditionalinterworking architecture is O(O − 1)/2. The number of SLAsrequired with the IISA architecture is O. When O is very high,IISA allows a significant reduction on the number of SLAs.

VI. PERFORMANCE EVALUATION

An analytical framework for evaluating the performanceof IPv6-based handoff schemes proposed by the IETF (i.e.,MIPv6, HMIPv6, FMIPv6, and F-HMIPv6) is presented in[33]. Such evaluation methods will be used to compare theperformance of the IETF’s protocols with HPIN. The parametervalues used in the performance evaluation are given in Table III,except when wireless link delay Lwl, packet arrival rate λp,prediction probability Ps, and user density in low-tier networksρl are considered variable parameters. The network topologyconsidered for analysis is illustrated in Fig. 8. We assume thedistance between different domains to be equal, i.e., c = d =e = f = 10 and set a = 1, b = 2, and g = 4. All links are sup-posed to be full duplex in terms of capacity and delay. Parame-ter values used to compute signaling cost are defined as follows:τ = 1, κ = 10, α = 0.2, β = 0.8, PCAEN = 8, PCHA = 24,


Fig. 9. BU signaling cost.

Fig. 10. Packet delivery cost versus packet arrival rate.

PCCN = 4, PCIDE = 15, and PCBEN = 12. The values ofother parameters are εl = εh = 10, Nl = 40, Nh = 5, Ns =15, νl = νh = 0.1, THA = TIDE = 20 min, PBU = 0.008 ms,PBR = 0.001 ms, PPU = 0.002 ms, and PPR = 0.005 ms.

Fig. 9 illustrates the BU signaling cost as a function of theSMR. When the SMR is small, the mobility rate is superiorto the session arrival rate, and the MN frequently changes itspoint of attachment, resulting in several handoffs. Then, thesignaling traffic overhead increases. The signaling overhead isconsiderably reduced from FMIPv6 to HPIN. However, whenthe session arrival rate is greater than the mobility rate (i.e.,SMR > 1), the BU is performed less often. In other words,signaling overhead decreases as the subnet change frequencydecreases. The HPIN enables significant cost saving in termsof signaling overhead. Additional messages introduced in theHPIN to allow handoff anticipation cause the signaling over-head to slightly increase compared to HMIPv6. However, thissignaling overhead increment is compensated by lower handofflatency and packet loss, as shown here. The packet deliverycost is depicted in Fig. 10 as a function of packet arrival

Fig. 11. Packet delivery cost versus prediction probability.

Fig. 12. Handoff latency versus wireless link delay.

rate λp. The HPIN outperforms all other IPv6-based handoffmanagement schemes, and the HPIN is more efficient when λp

increases. This means that the HPIN is highly suitable for real-time applications where periodic packets are sent at a high rate.

Fig. 11 shows the packet delivery cost for varying predictionprobability Ps. The packet delivery cost decreases as the accu-racy of Ps increases in fast handoff schemes. Higher Ps valuemeans that the FBAck message is received through the previouslink. Then, packets are delivered to an MN just after beingattached to the NAR/AEN. Results show that the HPIN per-forms better than all other schemes as it provides a lower packetdelivery cost. The prediction probability has a huge effect onF-HMIPv6, and if Ps = 0, F-HMIPv6 turns into HMIPv6,which is its reactive mode.

Fig. 12 shows that the handoff latency linearly increases withthe wireless link delay. MIPv6 has the worst performance com-pared to other schemes, followed by HMIPv6. Furthermore,FMIPv6 and F-HMIPv6 allow handoff latency reduction forMIPv6 and HMIPv6. In addition, the HPIN allows a significant


Fig. 13. Handoff latency versus prediction probability.

Fig. 14. Packet loss versus packet arrival rate.

handoff latency reduction compared to other mobility manage-ment protocols. It is well known that the maximal tolerabledelay for interactive conversation is approximately 200 ms.Hence, the HPIN meets this requirement when the wireless linkdelay is set below 60 ms. The effect of prediction probabilityPs on handoff latency is shown in Fig. 13. Regardless of the Ps

value, the HPIN performs better than all the other protocols.Fig. 14 shows the total packet loss in terms of packet arrival

rate. Packet loss values are much lower for the HPIN thanother IPv6-based handoff protocols. The effect of handoff inIPv6-based wireless environments is dominated by packet loss,which is due to the L2 handoff and the IP layer operations.In fact, due to the lack of buffering and anticipated handoffmechanisms in MIPv6 and HMIPv6, all in-flight packets arelost during handoff. However, in fast handoff schemes (i.e.,FMIPv6, F-HMIPv6, and HPIN), packet loss begins when L2handoff is detected until the buffering mechanism is initiatedor if the buffers overflow. Fig. 15 shows that the HPIN has

Fig. 15. Comparison of handoff-blocking probability.

Fig. 16. Processing load ratio versus the number of low-tier networks.

much lower handoff-blocking probability than other IPv6-based handoff schemes. This result is due to the ability of theHPIN to reduce signal message exchanges and handoff latency.Thus, HPIN can safely provide seamless handoff with servicecontinuity.

Fig. 16 shows the impact of the number of low-tier networkson the processing load for different values of the MN’s averagespeed. Results show that the IDE processing load is lower thanthat at the HA required for MIPv6. Thus, the IDE load due to in-tersystem and/or interdomain handoffs is limited. On the otherhand, one HA is usually used to handle MIPv6 handoff in theservice coverage area (e.g., one city) by network operators. Wecan thus conclude that a single IDE will be sufficient to handleintersystem and/or interdomain handoffs for the coverage areaof one city.

Fig. 17 illustrates that, as user density increases, the process-ing load for intersystem and/or interdomain handoffs at the IDEremains insignificant compared to the processing load at theHA for MIPv6. Fig. 18 shows that the IDE processing loadincreases as the number of cities increases. This means thatthe IDE load proportionally increases to the size of the service


Fig. 17. Ratio of processing load versus user density.

Fig. 18. Processing load at the IDE versus the number of cities.

coverage area. Therefore, an MN with a higher average veloc-ity is associated with a greater domain crossing rate, whichresults into a higher number of handoff requests. Such resultsencourage the deployment of the IDE through the hierarchicalarchitecture to allow the integration and the interworking ofvarious networks.

VII. CONCLUSION

Mobility management and system interworking are crucial in4G/NGWNs. Several IPv6-based protocols have been proposedfor mobility management at the IP layer. However, they cannotguarantee seamless roaming and service continuity for real-time applications. On the other hand, interworking architecturesavailable in the literature fail to fulfill all the requirements fordelay-sensitive and loss-sensitive applications.

To enable a better network performance in heterogeneousIP-based wireless and mobile environments, this paper hasproposed a novel interworking architecture called IISA and ahandoff management protocol called the HPIN. The proposedinterworking architecture IISA is based on an adaptive loose

coupling approach and introduces a third-party entity called theIDE to guarantee the seamless roaming and service continuityrequired in 4G/NGWNs. Moreover, IISA has several advan-tages such as scalability and easy deployment, and it supportsroaming between various heterogeneous wireless networks.

The HPIN is a one-suite protocol that carries out access net-work discovery, fast handoff, and localized mobility manage-ment. The HPIN reduces service disruption during a handoff byanticipating the handoff and allowing the selection of the bestavailable network. The performance analysis demonstrates sig-nificant gains for QoS defined in terms of signaling overhead,handoff latency, packet loss, and handoff-blocking probabilitythan current mobility management protocols. The IISA andHPIN can guarantee seamless handoff, service continuity, andQoS for an MN roaming across heterogeneous IP-based wire-less environments. Furthermore, the HPIN and IISA are simpleenough; thus, their deployment will not require strong effortand extensive costs. Future work includes validating numericalresults by using intensive simulations and prototypes.

ACKNOWLEDGMENT

The authors would like to thank the anonymous reviewers fortheir valuable comments that helped to improve the presentationof this paper.

REFERENCES

[1] S. Y. Hui and K. H. Yeung, “Challenges in the migration to 4G mobilesystems,” IEEE Commun. Mag., vol. 41, no. 12, pp. 54–59, Dec. 2003.

[2] M. Stemm and R. H. Katz, “Vertical handoffs in wireless overlaynetworks,” ACM Mobile Netw. Appl. (MONET), vol. 3, no. 4, pp. 335–350, Dec. 1998.

[3] E. Gustafsson and A. Jonsson, “Always best connected,” WirelessCommun., vol. 10, no. 1, pp. 49–55, Feb. 2003.

[4] I. F. Akyildiz, S. Mohanty, and J. Xie, “A ubiquitous mobile communi-cation architecture for next-generation heterogeneous wireless systems,”IEEE Commun. Mag., vol. 43, no. 6, pp. S29–S36, Jun. 2005.

[5] 3GPP System to WLAN Interworking; System Description (Release 6),Mar. 2004. 3GPP TS 23.234 v6.3.0.

[6] CDMA2000-WLAN Interworking; Stage 1 Requirements, Feb. 2006.3GPP TS, 3GPP2 S.R0087-A v1.0.

[7] J. McNair and F. Zhu, “Vertical handoffs in fourth-generation multi-network environments,” Wireless Commun., vol. 11, no. 3, pp. 8–15,Jun. 2004.

[8] D. B. Johnson, C. E. Perkins, and J. Arkko, Mobility Support in IPv6,Jun. 2004. IEFT RFC 3775.

[9] X. Pérez-Costa, M. Torrent-Moreno, and H. Hartenstein, “A performancecomparison of mobile IPv6, hierarchical mobile IPv6, fast handovers formobile IPv6 and their combination,” ACM SIGMOBILE Mobile Comput.Commun. Rev., vol. 7, no. 4, pp. 5–19, Oct. 2003.

[10] Y. Gwon, J. Kempf, and A. Yegin, “Scalability and robustness analysis ofmobile IPv6, fast mobile IPv6, hierarchical mobile IPv6, and hybrid IPv6mobility protocols using a large-scale simulation,” in Proc. IEEE ICC,Jun. 2004, vol. 7, pp. 4087–4091.

[11] H. Soliman, C. Castelluccia, K. El-Malki, and L. Bellier, Hierarchi-cal Mobile IPv6 Mobility Management (HMIPv6), Aug. 2005. IETFRFC 4140.

[12] G. Koodli, Fast Handovers for Mobile IPv6, Jul. 2005. IETF RFC 4068.[13] H. Y. Jung, E. A. Kim, J. W. Yi, and H. H. Lee, “A scheme for supporting

fast handover in hierarchical mobile IPv6 networks,” ETRI J., vol. 27,no. 6, pp. 798–801, Dec. 2005.

[14] M. Leibsch, A. Singh, H. Chaskar, D. Funato, and E. Shim, CandidateAccess Router Discovery (CARD), Jul. 2005. IETF RFC 4066.

[15] J. Loughney, M. Nakhjiri, C. Perkins, and R. Koodli, Context TransferProtocol (CXTP), Jul. 2005. IETF RFC 4067.

[16] Feasability Study on 3GPP System to WLAN Interworking (Release 6),Sep. 2003. 3GPP TS 22.934 v6.2.0.


[17] 3GPP2-WLAN Interworking; Stage 1 Requirements, Jul. 2004. 3GPP2TS, 3GPP2 S.R0087-0 v1.0.

[18] M. Buddhikot, G. Chandranmenon, S. Han, Y. W. Lee, S. Miller, andL. Salgarelli, “Integration of 802.11 and third-generation wireless datanetworks,” in Proc. IEEE INFOCOM, Apr. 2003, vol. 1, pp. 503–512.

[19] F. Du, L. M. Ni, and A.-H. Esfahanian, “HOPOVER: A new handoffprotocol for overlay networks,” in Proc. IEEE ICC, May 2002, vol. 5,pp. 3234–3239.

[20] W. Wang and I. F. Akyildiz, “A new signaling protocol for intersys-tem roaming in next-generation wireless systems,” IEEE J. Sel. AreasCommun., vol. 19, no. 10, pp. 2040–2052, Oct. 2001.

[21] R. Beaubrun, S. Pierre, and J. Conan, “An approach for managing globalmobility and roaming in the next-generation wireless system,” Comput.Commun., vol. 28, no. 5, pp. 571–581, Mar. 2005.

[22] H. J. Wang, R. H. Katz, and J. Giese, “Policy-enabled handoffs across het-erogeneous wireless networks,” in Proc. 2nd IEEE WMCSA, Feb. 1999,pp. 51–60.

[23] F. Zhu and J. McNair, “Multiservice vertical handoff decision algorithms,”EURASIP J. Wireless Commun. Netw., vol. 2006, 2006. 13 pages.

[24] C. Makaya and S. Pierre, “Handoff protocol for heterogeneous all-IP-based wireless networks,” in Proc. IEEE CCECE, Ottawa, ON, Canada,May 2006, pp. 223–226.

[25] S. Thomson and T. Narten, IPv6 Stateless Address Autoconfiguration,Dec. 1998. IETF RFC 2462.

[26] Y. Fang, “Movement-based mobility management and trade off analysisfor wireless mobile networks,” IEEE Trans. Comput., vol. 52, no. 6,pp. 791–803, Jun. 2003.

[27] W. Wang and I. F. Akyildiz, “Intersystem location update and pagingschemes for multitier wireless networks,” in Proc. ACM MOBICOM,Aug. 2000, pp. 99–109.

[28] F. V. Baumann and I. G. Niemegeers, “An evaluation of location man-agement procedures,” in Proc. 3rd Annu. Int. Conf. UPC, Sep./Oct. 1994,pp. 359–364.

[29] Y. Xiao, Y. Pan, and J. Lie, “Design and analysis of location managementfor 3G cellular networks,” IEEE Trans. Parallel Distrib. Syst., vol. 15,no. 4, pp. 339–349, Apr. 2004.

[30] J. Xie and I. F. Akyildiz, “A novel distributed dynamic location manage-ment scheme for minimizing signaling costs in mobile IP,” IEEE Trans.Mobile Comput., vol. 1, no. 3, pp. 163–175, Jul.–Sep. 2002.

[31] R. Koodli and C. E. Perkins, “Fast handovers and context transfers inmobile networks,” ACM SIGCOMM Comput. Commun. Rev., vol. 31,no. 5, pp. 37–47, Oct. 2001.

[32] J. McNair, I. F. Akyildiz, and M. D. Bender, “Handoffs for real-time trafficin mobile IP version 6 networks,” in Proc. IEEE GLOBECOM, Nov. 2001,vol. 6, pp. 3463–3467.

[33] C. Makaya and S. Pierre, “An analytical framework for performanceevaluation of IPv6-based mobility management protocols,” IEEE Trans.Wireless Commun., 2008, to be published.

Christian Makaya (S’06) received the M.Sc. degreein computer science from the University of Montreal,Montreal, QC, Canada, and the M.Sc. degree intelecommunications from the Institut National de laRecherche Scientifique–Centre Energie, Materiauxet Telecommunications (INRS-EMT), University ofQuebec, Montreal, in 2003 and 2004, respectively.He is currently working toward the Ph.D. degreein computer engineering at the Ecole Polytechniquede Montreal. His thesis is carried out jointly withEricsson Research Canada.

He is currently a Graduate Research Assistant with the Mobile Computingand Networking Research Laboratory (LARIM), Department of ComputerEngineering, Ecole Polytechnique de Montreal. His research interests includeradio resource management, interworking architecture design, mobility man-agement, IP mobility, performance evaluation, and quality-of-service provision-ing for next-generation wireless networks.

Samuel Pierre (S’90–M’91–SM’98) received theB.Eng. degree in civil engineering from the EcolePolytechnique de Montreal, Montreal, QC, Canada,in 1981, the B.Sc. and M.Sc. degrees in mathematicsand computer science from the Université du Québecà Montréal (UQAM), in 1984 and 1985, respectively,the M.Sc. degree in economics from the Universityof Montreal in 1987, and the Ph.D. degree in elec-trical engineering from the Ecole Polytechnique deMontreal in 1991.

From 1987 to 1998, he was a Professor with theUniversity of Québec, Trois-Rivières, QC, prior to joining the Télé-Universitéof Québec; an Adjunct Professor with the Université Laval, Sainte-Foy, QC;and an Invited Professor with the Swiss Federal Institute of Technology,Lausanne, Switzerland, and the Université Paris 7, Paris, France. He is cur-rently with the Ecole Polytechnique de Montreal, where he is a Professor ofcomputer engineering; the Director of the Mobile Computing and NetworkingResearch Laboratory (LARIM), Department of Computer Engineering; theNSERC/Ericsson Chair in Next Generation and Mobile Networking Systems;and the Director of the Mobile Computing and Networking Research Group(GRIM). He is a Regional Editor of the Journal of Computer Science andserves on the Editorial Board of Telematics and Informatics, which are edited byElsevier Science. His research interests include wireline and wireless networks,mobile computing, artificial intelligence, and telelearning.

Dr. Pierre is a Fellow of the Engineering Institute of Canada. He is an Asso-ciate Editor for the IEEE COMMUNICATIONS LETTERS, the IEEE CANADIAN

JOURNAL OF ELECTRICAL AND COMPUTER ENGINEERING, and the IEEECANADIAN REVIEW.

Date post:	05-Jan-2017
Category:	Documents
Upload:	trannguyet
View:	227 times
Download:	0 times

ieee transactions on vehicular

Documents