GUARANTEEING host mobility with quality of service

Transcription

1 260 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 1, JANUARY 2007 Optimization of Mobile IPv6-Based Handovers to Support VoIP Services in Wireless Heterogeneous Networks Hanane Fathi, Member, IEEE, Shyam S. Chakraborty, Member, IEEE, and Ramjee Prasad, Senior Member, IEEE Abstract The support of voice over Internet Protocol (VoIP) services in next-generation wireless systems requires the coupling of mobility with quality of service. The mobile node can experience disruptions or even intermittent disconnections of an ongoing real-time session due to handovers. The duration of such interruptions is called disruption time or handover delay and can heavily affect user satisfaction. Therefore, this delay needs to be minimized to provide good-quality VoIP services. In this paper, the focus is on the network layer mobility, specifically on mobile Internet Protocols (MIPs), since they are natural candidates for providing mobility at layer 3. Using analytical models, the authors evaluate MIPv4, MIPv6, fast MIPv6 (FMIPv6), and hierarchical MIPv6 (HMIPv6) and compare their performances in terms of handover delay for VoIP services. To optimize the handover delay, the authors propose to use the adaptive retransmission timer described in this paper. The results obtained using the adaptive timer technique show that for a 3% frame error rate and a 128-kb/s channel, the handoff delay is about s (predictive) and s (reactive) for FMIPv6. It is around s [intra-mobile anchor point (MAP)] and 1.47 s (inter-map) for HMIPv6, around 1 s for MIPv6, and 0.26 s for MIPv4. Index Terms Fast mobile IPv6 (FMIPv6), handover delay, hierarchical mobile IPv6 (HMIPv6), Internet Protocol (IP)-based wireless networks, mobile IPv6 (MIPv6), voice over IP (VoIP). I. INTRODUCTION GUARANTEEING host mobility with quality of service is one of the main challenges in heterogeneous wireless systems. As voice over Internet Protocol (VoIP) communications increase and are being extended to mobile networks, it is important to support user and host mobility with a satisfactory level of quality in the voice session. Host mobility is becoming essential because of the strong need to have continuous network connectivity. Moving from one place to another can be modeled Manuscript received September 30, 2004; revised August 30, 2005, December 20, 2005, and February 27, This work was supported in part by the Danish Statens Teknisk-Videnskabelige Forskningsrad through the Center for Network and Service Convergence (CNTK) and by the Academy of Finland. The review of this paper was coordinated by Dr. Q. Zhang. H. Fathi was with the Center for TeleInFrastuktur, Aalborg University, 9220 Aalborg, Denmark. She is now with the AIST Research Center for Information Security, Tokyo , Japan ( hanane.fathi@aist.go.jp). S. S. Chakraborty was with the Academy of Finland and the Helsinki University of Technology, Espoo, Finland. He is now with Ericsson Finland, Jorvas, Finland ( Shyam.chakraborty@ericsson.com). R. Prasad is with the Center for TeleInFrastruktur, Aalborg University, 9220 Aalborg, Denmark ( prasad@kom.aau.dk). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /TVT as changing the node s point of attachment to the network. Supporting mobility at the network layer is therefore naturally modeled as changing the routing of datagrams destined for the mobile node (MN), so that they arrive at the new point of attachment. In this paper, we focus on the mobility at a convergent layer for heterogeneous networks: network layer with mobile Internet Protocols (MIPs). The goal of the MIP [1] is to provide a host the ability to stay connected to the Internet regardless of its location. In MIPv4, the MN obtains a new IP address from a foreign router [foreign agent (FA)] in the visited network or through some external assignment mechanism and registers with the FA. To maintain continuous connectivity, the MN needs to update its location with its home agent (HA) whenever it moves to a new subnet so that the HA can forward it the packets. However, MIP is not a good solution for users with high mobility: It suffers from extra delays due to the routing of each packet through the HA (triangular routing), lack of addresses, and high signaling load. To overcome these issues, IETF has proposed MIPv6 [2], hierarchical MIPv6 (HMIPv6) [3], and fast MIPv6 (FMIPv6) [4]. During the handover process of any MIP-based protocol, the MN cannot receive IP packets on its new point of attachment until the handover ends. This can result in disruption of the ongoing media session and the dissatisfaction of the user. Therefore, it is important to evaluate the disruption time or handover delay of the mobility protocols. The disruption time or handover delay is defined here as the time interval from when the handover process starts to when the MN can send and receive data packets. The support of VoIP in mobile systems requires low handover latency (i.e., < 400 ms [5]) to achieve seamless handovers. Mobility management is the focus of much recent research. Through analytical results in [6] and [7] and simulations in [8], the seamless mobility issue is addressed by comparing MIPv4 and Session Initiation Protocol (SIP) mobility and proposing mechanisms to reduce the disruption time. In [9], SIP handoff delay is evaluated to be around 6 s for moderate frame error rate (FER) using an analytical model similar to the one used here. However, SIP handoff delay is too high to support real-time sessions. Therefore, it is not of interest to us to investigate SIP mobility in this paper. Also, an experimental study of MIPv4 performances over WLAN has been conducted in [10]. Concerning HMIPv6, in [11], a performance study has been done through simulations with Network Simulator 2, and in [12], /$ IEEE

2 FATHI et al.: OPTIMIZATION OF MOBILE IPv6 TO SUPPORT VoIP SERVICES IN WIRELESS NETWORKS 261 a comparison of micromobility protocols has been performed based on simulations with Columbia IP Micromobility Software. In [13] and [14], MIPv6 and its enhancements are compared in terms of scalability and robustness to failures using simulations. Our work differs from others in the following ways: 1) We evaluate the handover delay of the four MIPbased protocols using two analytical methods, and we compare their performances and robustness to channel errors in various conditions. Handover delays are analytically derived for each scheme in various situations. The first method gives a good orientation toward the crucial factors to be investigated further in the second method, namely the impact of the wireless link quality (i.e., the FER) on the handoff delay. The second method involves the reliability mechanism of each protocol to overcome losses that are most likely to happen over a wireless link. 2) We propose to use an adaptive retransmission timer to optimize the handover delay and compare the results obtained using the fixed timers specified in [1] [3] with our adaptive approach. To the best of our knowledge, such analysis and optimization for MIP-based protocols are unprecedented in the previous literature. The rest of the paper is organized as follows: In Sections II and III, the MIP-based protocols and their reliability mechanisms are described with their respective signaling flows. The performance analysis of the different schemes in terms of handover delay using two analytical models and the adaptive retransmission timer is given in Section IV. Then, the results are presented in Section V, considering various conditions, and the concluding remarks are given in the last section. II. OVERVIEW OF MIP-BASED PROTOCOLS A. MIPv4 The MIP [1] process has three main mechanisms, namely 1) agent discovery, 2) registration, and 3) tunneling. HA and FA advertise their presence via agent advertisement messages so that they become known by the MN. An MN may optionally solicit an agent advertisement message from any locally attached agent through an agent solicitation message and receives the agent advertisements. Then, it determines whether it is on its home network or on a foreign network. When an MN detects that it has moved to a foreign network, it obtains a care-ofaddress (CoA) on the foreign network. The CoA is the endpoint of a tunnel toward an MN to receive the packets forwarded by the HA while it is away from home. Tunneling is the method used to forward the message from the HA to the FA and, finally, to the MN by encapsulating the original message in a new IP packet containing the header CoA as the destination address. There are two different types of CoAs: a Foreign Agent CoA that is the address of the FA with which the MN is registered (the tunnel ends at the FA); a Collocated CoA that is an externally obtained address which the MN has associated with one of its interfaces (the tunnel is established directly from the HA to MN). After obtaining a CoA, the MN registers its CoA with its HA to obtain service. This handover process is illustrated in Fig. 1. Packets sent to the MN s home address are intercepted and tunneled to the MN s CoA by the HA, received Fig. 1. MIPv4 handover process. at the tunnel endpoint (either at an FA or at the MN itself for collocated CoA), and finally delivered to the MN at its current location. In the reverse direction, datagrams sent by the MN are generally delivered to their destination using standard IP routing mechanisms. This can result in incremental delays of the packets. To address this problem, route optimization [15] has been defined and requires the HA to send to any CN the MN s current CoA. Therefore, the CNs can send datagrams directly to a mobile node, and the delay due to the triangular routing is minimized. B. MIPv6 MIPv6 shares many features with MIPv4, but it is integrated into IPv6 and offers many other improvements. IPv6 introduces 16 bytes-length addresses that are autoconfigured at each IPv6 node. Thus, there is no need for any FAs in IPv6 mobility support. A binding update (BU) option is defined for mobility support that combines the functions of the registration request for IPv4 and the BU message for route optimization. Therefore, the MN entering in a foreign domain updates its location at the HA and at the CN by exchanging BUs/binding acknowledgments (BAs) with both entities. Also, to ensure a secure BU at the CN, Johnson et al. [2] have defined a new method called the return routability procedure. The basic return routability mechanism consists of two checks, namely 1) a home address check and 2) a CoA check to guarantee the legitimacy of the MN. This procedure consists of the exchange of four messages with CN prior to sending the BU messages. The MN sends to the CN two messages at the same time: Home Test Init message via the HA and Care-of Test Init message directly. Upon the reception of each message, the CN sends back two messages to the MN: Home Test message via the HA and Care-of Test message directly, each containing a different token to be used by the MN to generate the binding management key. This binding management key is then used by the MN to send a verifiable BU to the CN. The handover process is illustrated in Fig. 2. Existing bindings become obsolete each time the MN moves to a new point of attachment and autoconfigures a CoA. When this happens, the MN should immediately send out BUs to all correspondents with which it is actively communicating. C. HMIPv6 HMIPv6 is an enhancement of MIPv6 protocol, which aims to reduce the amount of signaling protocol required and

3 262 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 1, JANUARY 2007 Fig. 4. HMIPv6 intra-map handover process. Fig. 2. MIPv6 handover process. Fig. 5. FMIPv6 predictive handover process. MAP domain (LCoA), it only needs to register the new address with the MAP, as shown in Fig. 4. The RCoA must be registered with CN and HA only when the MN moves outside a MAP domain, as shown in Fig. 3. Fig. 3. HMIPv6 inter-map handover process. improving handoff delays for mobile connections. Although it is not necessary for external hosts to be updated when an MN moves locally, these updates occur for both local and global movements. To solve this inefficient use of resources, in the case of local mobility, HMIPv6 adds another level on MIPv6, separating local mobility from global mobility. HMIPv6 introduces a new entity called mobile anchor point (MAP). The MAP replaces the MIPv4 FA and helps to decrease handover latency because a local MAP can be updated more quickly than a remote HA. The MAP can be located anywhere within the architecture of routers. As illustrated in Fig. 3, an MN entering a MAP domain receives router advertisements containing information on one or more local MAPs. The MN can bind its current location [on-link CoA (LCoA)] with an address on the MAP s subnet [regional CoA (RCoA)]. Acting as a local HA, the MAP receives all packets on behalf of the MN it is serving and encapsulates and forwards them directly to the MN s current address. If the MN changes its current address within a local D. FMIPv6 FMIPv6 [4] is another enhancement of MIPv6, which aims to reduce handoff delays for mobile connections by delivering the packet in the new point of attachment at the earliest. There are two modes of operations: predictive and reactive. In both modes, the MN sends a router solicitation for proxy advertisement (RtSolPr) to its current access router (AR). The AR replies with a proxy router advertisement (PrRtAdv) that provides to the MN information about the neighboring AR so that the MN can formulate a prospective new CoA. Then, the MN sends a fast BU (FBU) that allows the previous AR to tunnel packets destined to the MN from the old CoA to the new CoA. In the predictive mode illustrated in Fig. 5, the FBU is sent from the link with the previous AR. The previous AR checks with the new AR as to whether the new CoA is acceptable by exchanging handover initiate (HI) and handover acknowledge (HAck). The MN then receives a fast binding acknowledgment (FBAck) that informs it that the tunneling is in progress. The MN should then immediately send a fast neighbor advertisement (FNA) after attaching to the new AR so that the new AR can start delivering buffered packets to the MN. In the reactive mode illustrated in Fig. 6, the MN has already moved to the new AR and did not receive an FBAck. It thus sends an FBU encapsulated in an FNA via the new AR to the previous AR, which sends back an FBAck to the new AR and starts forwarding packets to the new AR if the new CoA is accepted. Then, the new AR delivers them immediately to the MN.

4 FATHI et al.: OPTIMIZATION OF MOBILE IPv6 TO SUPPORT VoIP SERVICES IN WIRELESS NETWORKS 263 Fig. 6. FMIPv6 reactive handover process. III. RELIABILITY MECHANISMS FOR MIP-BASED PROTOCOLS In this section, we describe the retransmission mechanisms needed in MIPv4, MIPv6, HMIPv6, and FMIPv6 handover procedures to overcome losses that are likely to happen on the wireless link. A. MIPv4 MIPv4 handoff involves two procedures, namely agent discovery and registration procedure. Each procedure has a specific retransmission mechanism. For agent discovery, the rate at which an MN sends solicitations must be limited. The MN sends three initial solicitations at a maximum rate of 1/s while searching for an agent. After sending the three initial solicitations, the rate is reduced to limit the overhead on the local link. Subsequent solicitations are sent using a binary exponential back-off mechanism, doubling the interval between consecutive solicitations, up to a maximum interval. This maximum interval is 1 min between solicitations [1]. For the registration procedure, if no registration reply has been received within a reasonable time, another registration request is transmitted. The maximum time until a new registration request is sent is not greater than the requested lifetime of the registration request. If the MN is registering with an FA, the lifetime should not exceed the value in the registration lifetime field of the agent advertisement message received from the FA. The minimum value should be large enough to account for the size of the messages, twice the round-trip time for transmission to the HA, and at least an additional 100 ms, to allow for processing of the messages before responding. The round-trip time for transmission to the HA is at least as large as the time required to transmit the messages at the link speed of the MN s current point of attachment. The minimum time between registration requests is not less than 1 s. Each successive retransmission timeout period is at least twice the previous period, as long as it is less than the maximum [1]. B. MIPv6, HMIPv6, and FMIPv6 As in MIPv4, router discovery and registration procedures have different retransmission mechanisms. For the router discovery, we assume that the router solicitation and the router advertisement retransmission mechanisms are the same as the one used in MIPv4 for agent solicitation and agent advertisement. For the registration, the MN is responsible for the retransmissions and the rate limit for the registrations. When the MN sends a BU for which it expects a response, the MN has to determine a value for the initial retransmission timer. If the MN sends a BU without having an existing binding at the HA, it uses InitialBindackTimeoutFirstReg (1.5 s) as the initial retransmission timer value. This long retransmission interval allows the HA to complete the duplicate address detection (DAD) procedure, which is mandated in this case. Otherwise, the MN uses the specified value of INITIAL_BINDACK_TIMEOUT (1 s) for the initial retransmission timer. If the MN fails to receive a valid matching response within the selected initial retransmission interval, the MN retransmits the message until a response is received. The retransmissions follow an exponential back-off process, in which the timeout period is doubled upon each retransmission until either the node receives a response or the timeout period reaches 32 s [2]. The return routability procedure follows the same mechanism as the registration. IV. PERFORMANCE ANALYSIS In this section, we analyze the handover delay of the original MIPv4, MIPv6, HMIPv6, and FMIPv6. In this paper, we evaluate the time interval between the moment when the MN sends an agent/router solicitation and the moment when the MN can send and receive IP packets to/from the CN, under various conditions. We focus here on network layer handover; therefore, we do not consider link layer detection. It important to note that registration can also be caused by expiration of bindings at the HA and at the CNs. BUs set up a certain lifetime for the bindings; thus, even if the MN does not change its point of attachment, the MN should send BUs to its HA and to the CNs before the lifetime expires to keep its bindings active. The analysis consists of two steps. 1) The first step is based on a simple model presented in [6] that takes into account the delay increases between the different entities involved in the handover. 2) The second step considers the FER of the wireless link and the retransmissions strategies of the different protocols to overcome the losses. A. Step 1 For simplicity, we consider the model illustrated in Fig. 7. The following notations are used. The delay between the MN and the radio access network (RAN) is t mr, which is the time to send a message over the subnet via wireless link. The delay between the MN and the AR is t s. The delay between the previous AR and the new AR is t no. The delay between the MN and the FA/MAP is t mf.

5 264 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 1, JANUARY 2007 Fig. 7. Simple model for analysis. The delay between the MN and its HA is assumed to be t h, which is the time necessary for a message to be delivered to the home network. The delay between the MN and the CN is t mc. The delay between the MN s home network and the CN is t hc. We make the following assumptions. The delays are considered symmetric. t s <t h. In MIPv4, we use FA CoA instead of colocated CoA; therefore, the MN s incoming and outgoing traffic is relayed by the FA (when the MN is in a foreign network). Solicitations are needed to be sent to receive advertisements. For each BU message sent, a binding acknowledge (BA) is expected to be received. The bit A of the BU message is put to 1 to request the BA. For MIPv6 and HMIPv6 registrations, we do not consider the time needed by DAD process. The processing and queuing times are not considered here. 1) MIPv4 Handoff: In MIPv4, the handoff is handled as follows: The MN detects the IP subnet by exchanging agent solicitation and agent advertisement messages, which takes 2t mf. Then, the MN sends a MIP registration request to the HA and gets a registration reply, which takes 2t h. At this point, the MN starts receiving downlink packets. The MIPv4 handoff takes 2t mf +2t h. 2) MIPv6 Handoff: In MIPv6, the MN detects the IP subnet by exchanging with AR router solicitation and router advertisement messages, which takes 2t s. Then, the MN sends to the HA a BU and gets a BA, which takes 2t h. Then, Home and Care-of Test Init messages are sent by the MN at the same time. The procedure requires very little processing at the CN, and the Home and Care-of Test messages can be returned quickly (nearly simultaneously) [2]. As MN can only generate a verifiable BU once it received both Test messages, we have to consider the longest of both exchanges, i.e., the one via the HA. Therefore, the delay to perform the return routability is equal to the exchange of the Home Test Init and Home Test messages, which takes 2t h +2t hc. Finally, the MN sends to the CN a BU and gets a BA, which takes 2t mc. The MIPv6 handoff delay is 2t s +4t h +2t mc +2t hc. 3) HMIPv6 Handoff: For HMIP, two handover cases have to be distinguished: 1) the inter-map handoff and 2) the intra- MAP handoff. For the intra-map handoff, the MN detects the IP subnet by exchanging with AR router solicitation and router advertisement messages, which takes 2t s. Then, the MN sends to the MAP a BU and gets a BA, which takes 2t mf. The intra- MAP handoff delay is 2t s +2t mf. For inter-map handoff, the procedure starts in the same way as for intra-map handoff with router solicitation/advertisement and BU/BA to the MAP, which takes 2t s +2t mf.themn also needs to update the binding cache at the HA and the CN by performing the return routability procedure and exchanging verifiable BU/BA, which takes 4t h +2t mc +2t hc. The HMIPv6 handoff delay is 2t s +2t mf +4t h +2t mc +2t hc for inter-map mobility. 4) FMIPv6 Handoff: For FMIPv6, two handover cases have to be distinguished: the predictive handoff and the reactive handoff. In the predictive handoff, the MN obtains a new CoA after exchanging RtSolPr and PrRtAdv with the previous AR, which takes 2t s. The MN sends an FBU to the previous AR, which takes t s. The ARs then exchange HI and HAck, which takes 2t no. The previous AR sends an FBAck to the new AR and to the MN, which takes at most t s. Finally, the MN sends an FNA to the new AR, which takes t s. The predictive FMIPv6 handoff takes 5t s +2t no. In the reactive handoff, the MN obtains a new CoA after exchanging RtSolPr and PrRtAdv with the previous AR, which takes 2t s. The MN sends an FBU encapsulated in an FNA to the previous AR, which takes t s + t no. The ARs then exchange FBU and FBAck, which takes 2t no. The reactive FMIPv6 handoff takes 3t s +3t no. B. Step 2 In this section, we evaluate the handoff delay as a function of the FER. We make the following assumptions. There is a random error process. The link layer reliability mechanism operates in the unacknowledge mode.

6 FATHI et al.: OPTIMIZATION OF MOBILE IPv6 TO SUPPORT VoIP SERVICES IN WIRELESS NETWORKS 265 TABLE I BACK-OFF TIMERS SPECIFIED IN [1] [3] the solicitation/advertisement transaction: transmission of the agent solicitation (containing k 1 frames), acknowledged by the agent advertisement (containing k 2 frames). Hence, the retransmission timer of this transaction is Tr(1) = D +(k 1 1) τ + D +(k 2 1) τ +2 t rf (2) where t rf is the delay between the RAN and the FA (t rf = t mf t mr ).ThevalueofTr(1) is changing, reflecting the size of the messages exchanged in the transaction and the path taken by the messages. 3) Retransmission Probability: The probability of retransmission q is the probability of a transaction having failed: This means that the first packet sent (solicitation containing k 1 frames) is lost or that the first packet is received but that the response (advertisement containing k 2 frames) is lost. Therefore, the probability of having a retransmission of solicitation is q = ( 1 (1 p) k ) ( ) ( ) (1 p) k 2 (1 p) k 1 (3) An agent/router advertisement is sent only if an agent/router solicitation has been previously received. A registration reply/ba is sent only if a registration request/bu has been previously received. Error correcting codes, processing, and queuing times are not considered here. Let p be the probability of a frame being erroneous in the air link. Therefore, considering k frames contained in a packet, the packet loss rate is (1 (1 p) k ). We denote τ as the interframe time, which is the time interval between the transmissions of two consecutive frames, and D as the frame propagation delay through the RAN. Therefore, the propagation delay from MN to RAN for a MIP message is D +(k 1)τ. 1) Exponential Back-Off Mechanism: The retransmission timers for all the MIP-based protocols follow the exponential back-off mechanism. Let Tr(1) be the initial back-off timer. The back-off timer upon the ith transmissiontr(i) doubles after each retransmission. Hence Tr(i) =2 i 1 Tr(1). (1) The upper bound for Tr(i) is 2 Nm 1 Tr(1), where Nm is the maximum number of transmissions allowed. 2) Adaptive Retransmission Timer: The initial retransmission timer Tr(1) can be taken from the specification (see Table I). However, the initial retransmission timer Tr(1) is a crucial parameter that should be optimized since it is of a direct impact on the handover delay. It should not be too short; otherwise, a packet is retransmitted while a response is on the way to being received. It should not be too long in order to avoid increasing the handover delay unnecessarily if a loss occurs. Therefore, it has to be proportional to the transmission time of the messages involved in the handover transaction. We define a transaction as the exchange of solicitation/advertisement or request/reply messages. It is function of the number of frames k contained in the MIP messages of the frame RAN propagation delay D and of the interframe time τ. Let us consider q =1 ( (1 p) k 1+k 2 ). (4) The value of q is changing, reflecting the size of the messages exchanged in the transaction. 4) Normalized Handover Delay: Let Nm be the maximum number of transmissions. The normalized delay for a successful transaction is the normalized delay for successfully transmitting a MIP message and successfully receiving the corresponding acknowledgment. This is because the sender knows that its sent packet (e.g., solicitation) has successfully been received when it receives an acknowledgment (e.g., advertisement). Therefore, the normalized delay Tt(i) MIP for the successful transmission of the ith MIP request message is as follows: Tt(i) MIP = 1 1 q Nm [(1 q)(d +(k 1)τ) +(1 q)q (Tr(1) + D +(k 1)τ) +(1 q)q 2 (3Tr(1) + D +(k 1)τ) + +(1 q)q Nm 1 ((2 Nm 1 1)Tr(1) + D +(k 1)τ )] = D +(k 1)τ Tr(1) + (1 q)(1 (2q)Nm ) (1 q Nm Tr(1). (5) )(1 2q) We assume that the transmission of the (i +1)th message (i.e., reply/advertisement/acknowledgment) depends on the successful reception of the ith message (i.e., request/ solicitation). Thus, the delay for transmitting reply/ advertisement is Tt(i +1)=D +(k 1)τ. The handover delay is the addition of the delays for all the N messages necessary to perform the handover. The normalized handover delay Tt MIP is given as Tt MIP = N Tt(i) MIP. (6) i=1

7 266 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 1, JANUARY 2007 For MIPv4, 1 the handoff delay is as follows: Tt MIPv4 = Tt(AgSol)+Tt(AgAdv) + Tt(RegReq)+Tt(RegRep) +2t rh +2t rf (8) where t rf is the delay between the RAN and the FA (t rf = t mf t mr ), and t rh is the delay between the RAN and the HA (t rh = t h t mr ). For MIPv6, the handoff delay is as follows: Tt MIPv6 = Tt(RSol)+Tt(RAdv) +2Tt(BU)+2Tt(BA) + Tt(HoTi)+Tt(HoT) +4t rh +2t rc +2t hc (9) where t rh is the delay between the RAN and the HA (t rh = t h t mr ), and t rc is the delay between the RAN and the CN (t rc = t mc t mr ). For HMIPv6 inter-map, the handoff delay is as follows: Tt inter HMIPv6 = Tt(RSol)+ Tt(RAdv) +3Tt(BU)+3Tt(BA) + Tt(HoTi)+Tt(HoT) +2t ra +2t rmap +4t rh +2t rc +2t hc (10) where t ra is the delay between the RAN (t ra = t s t mr ) and the AR, and t rmap is the delay between the RAN and the MAP (t rmap = t mf t mr ). For HMIPv6 intra-map, the handoff delay is as follows: Tt intra HMIPv6 = Tt(RSol)+Tt(RAdv)+2t ra +2t rmap + Tt(BU)+Tt(BA). (11) For FMIPv6 in the predictive mode, the handoff delay is as follows: Tt Pred FMIPv6 = Tt(RtSolPr)+ Tt(PrRtAdv) + Tt(FBU)+Tt(FBA) + Tt(FNA)+5t ra +2t no (12) where t no is the delay between the previous AR and the newar. 1 For solicitations, the retransmission mechanism specifies that the three initial transmissions are sent with 1-s interval, and then, the transmissions intervals follow the exponential back-off mechanism. Therefore Tt(AgSol) MIPv4 = 1 1 q Nm [(1 q)(d +(k 1)τ) +(1 q)q (Tr(1) + D +(k 1)τ) +(1 q)q 2 (2Tr(1)+D+(k 1)τ)+ +(1 q)q Nm 1 ((2 Nm 1 1)Tr(1) + Tr(1) + D +(k 1)τ )]. (7) It is similar for router solicitations in MIPv6. Fig. 8. Disruption time versus delay between MN and CN. For FMIPv6 in the reactive mode, the handoff delay is as follows: Tt React FMIPv6 = Tt(RtSolPr)+ Tt(PrRtAdv) + Tt(FNA)+3t ra +2t no. (13) V. N UMERICAL RESULTS In this section, we present results based on the previous analysis. The results from the two steps are given. A. Step 1 To evaluate the disruption time, we set t mr =10 ms as in [6], considering a relative low bandwidth in the wireless link. Also, we assume t mf = t mr +2=12 ms, t no =5 ms, and t s =11ms. The delay introduced by the Internet depends on the number of routers and the type of links in the path of datagram transmission. It is rather difficult to standardize such heterogeneous transmission paths and compute the transmission delay. For this reason, we have assumed the one-way Internet delay over the wired network to be constant, i.e., equal to 100 ms. Therefore, we assume t h = 112 ms, t mc = 124 ms, and t hc = 114 ms. 1) Impact of the Delay Between MN and CN: In Fig. 8, we can notice that the disruption times for MIPv4, FMIPv6, and HMIPv6 (intra-map) are independent of the delay increase between MN and CN. However, for MIPv6 and HMIPv6 (inter-map), the disruption time gets greater (up to 900 ms), while this delay increases (up to 150 ms). These results corroborate the intention of the protocol specifications: The MN in a foreign network only registers with its HA (MIPv4), with the ARs (FMIPv6), and with its MAP (HIMPv6 intra-map). On the other hand, for MIPv6 and HMIPv6 (inter-map), the MN registers with its CN, involving then an extra message to perform the return routability and registration procedure. Also, HMIPv6 (intra-map) gives the shortest handoff delay because the MN registers only to its MAP, which is in the same foreign network.

8 FATHI et al.: OPTIMIZATION OF MOBILE IPv6 TO SUPPORT VoIP SERVICES IN WIRELESS NETWORKS 267 TABLE II COMPARISON HANDOFF DELAY FOR FER =3%FOR A 128-kb/s CHANNEL Fig. 9. Disruption time versus delay MN and its home network. Fig. 10. Disruption time versus wireless link delay. 2) Impact of the Delay between MN and Its Home Network: In Fig. 9, the handoff delays with HMIPv6 protocol (intra- MAP) and with FMIPv6 are the lowest and independent of the increase of the delay between the MN and the home network because when the MN moves within a MAP domain, there is no need to update the RCoA with its HA and CN, and FMIPv6 does not involve home registrations. On the other hand, the HA is involved in the registration process for the other protocols analyzed, resulting in higher handoff delays when the delay between MN and its home network increases. MIPv6 and HMIPv6 for inter-map movements result in a handover delay unacceptable for VoIP sessions, even if the delay between the MN and its home network is low. This is also due to the return routability procedure. MIPv4 handover delay is at the borderline for providing an undisturbed session with 220 ms for a 120-ms delay between the MN and its home network. 3) Impact of the Wireless Link Delay: The results obtained with varying wireless link delay give more insight to the comparisons. Indeed, Fig. 10 shows that the wireless link delay increase affects the handoff delay for all the protocols analyzed. The HMIPv6 (inter-map) case is the most affected because it involves the highest number of messages exchanged over the wireless interface. However, when an MN moves within a MAP domain (HMIPv6 intra-map), the handoff delay is the least affected by the wireless link delay increase. HMIPv6 (intra-map) and MIPv4 handover delays increase with same slope with respect to the wireless link delay. FMIPv6 handover delays give the lowest slopes and the lowest values. In this case, FMIPv6 and HMIPv6 (intra-map) appear to be the most suitable approach for VoIP communications. Three first conclusions can be drawn from Step 1. The essential factors impacting on the handoff delay are the number of exchanged messages necessary to perform the handover and the entities involved in the process (i.e., HA, CN, MAP, FA, and AR). Local mobility management and fast handoff with prediction and reaction seem to give lower handover delay than the other MIP-based protocols analyzed. In the case of global handovers, the hierarchical management gives a higher handoff delay than for nonhierarchical protocols. However, note that the MN is expected to move more often within the MAP domain than between two different MAPs, especially during a VoIP session. The wireless link delay has an important influence in the handoff delay and is investigated further in Step 2 with the FER model.

9 268 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 1, JANUARY 2007 Fig. 11. Disruption time versus FER with the fixed specified timers. Fig. 13. Disruption time versus FER with the proposed adaptive timers. TABLE III MESSAGE SIZES Fig. 12. Disruption time versus FER with the proposed adaptive timers. B. Step 2 The number of transactions and the size of the messages exchanged affect the handover delay. For the evaluation, the size of each message (See Table III) and the values of the fixed back-off timers are obtained from [1] [3], and those for adaptive timer are obtained from the transmission delay of the different transactions using (2). We consider a 128-kb/s channel. The values of the delay D and the interframe time τ are set, respectively, to 10 ms, as in [6], and to 1 ms. The number of maximum transmissions allowed (Nm) is 6. The handoff delay is evaluated at various FERs between 0% and 10%. However, the session is established for VoIP, and voice services are supported if the FER is between 1% and 3%. Results for an FER of 3% are presented in Table II using adaptive and fixed timers (cf. Table I). Comparing the results with Step 1, one can see that the protocols that show high delays are the same, i.e., HMIPv6 inter- MAP and MIPv6, because of the high number of exchanged messages over the wireless link. However, in contrast with Step 1, if fixed timers are used, MIPv4 shows the lowest delay for FER > 8% because another important parameter enters into account, i.e., the back-off timer (1 s), which is the lowest for registration request among all protocols (Table I). MIPv6, FMIPv6, and HMIPv6 have their BU timers set to 1.5 s. For FER < 8%, FMIPv6 reactive handoff gives the lowest delay as shown in Fig. 11. If adaptive timers are used, as shown in Figs. 12 and 13, the ranking of the different protocols is as in Step 1 (i.e., from best to worst performance: HMIPv6 intra-map, reactive FMIPv6, predictive FMIPv6, MIPv4, MIPv6, and, finally, HMIPv6 inter- MAP). However, the handover delay is significantly reduced: up to 55% shorter for HMIPv6 (inter-map), up to 50% shorter for MIPv6, up to 50% shorter for MIPv4, and up to 99% shorter for FMIPv6 and HMIPv6 (intra-map). Also, the difference between MIPv6 and HMIPv6 inter-map is bigger in proportion: HMIPv6 inter-map handover delay is around 40% higher than that of MIPv6. The adaptive timer brings down the disruption times at 3% FER for MIPv4 and HMIPv6 (intra-map) to a value that is acceptable for voice communications. The value of k is another parameter that influences the results. It depends a lot on the data rate considered and the number of frames contained in one packet sent over the air link. The greater this number is, the higher the probability of a packet being lost will be, and the longer the adaptive timer. Local handover management as in FMIPv6 and HMIPv6 seems to be appropriate for midsession mobility or midsession periodic location update. Even if the local mobility management implies that global management may happen during interdomain handovers and this provides extra delays than

10 FATHI et al.: OPTIMIZATION OF MOBILE IPv6 TO SUPPORT VoIP SERVICES IN WIRELESS NETWORKS 269 the classical management (i.e., in MIPv4 and MIPv6), most midsession handovers or periodic updates take place within a MAP domain. In fact, the updates of the binding caches at the CN, MAP, and HA, as well as the setup of the tunnels MN-HA and MN-MAP, are necessary during MIP-based handovers, but they lengthen the handovers to various degrees for the three protocols. Although this hierarchical management of mobility has the drawback of a greater disruption time, it has the advantage of avoiding the extra delay introduced by the triangular routing and the extra overhead to the end-to-end VoIP messages due to the necessary encapsulation. VI. CONCLUSION In this paper, we have focused on mobility management issues at the network layer to support VoIP services in wireless heterogeneous networks. We first described briefly MIP-based protocols and compared their performances in terms of disruption time. To optimize handover delays, we have proposed using an adaptive retransmission timer, which is proportional to the size of the messages involved in the transactions of the handover process. Our analysis led to the following conclusion: Local mobility management as in FMIPv6 and HMIPv6 seems to be the most suitable approach to handle network layer mobility for VoIP. In most conditions, it provides minimum disruption for inter-ar movements, which are the most expected cases, and avoids triangular routing, which harms VoIP services in mobile systems. We have also identified four crucial parameters that affect the handover performances of the protocols, depending on the FER in the air link. These are the number of messages exchanged over the air link, the entities involved in the process, the retransmission strategy (maximum number of retransmissions allowed, back-off mechanism, and back-off timer), and the message sizes of the protocols. Reducing the back-off timer to an appropriate value via the adaptive back-off timer proposed in this paper leads to a drastic decrease of the handover delay. In the future, more investigations can be done on the impact of correlated errors on the disruption time, as well as the consideration of link layer retransmissions, error correction mechanisms, processing, and queuing delays. REFERENCES [1] C. Perkins, IP Mobility Support, Aug IETF RFC3344. [2] D. Johnson, C. Perkins, and J. Arkko, Mobility Support in IPv6, Jun IETF RFC [3] H. S. Flarion, C. Castellucia, K. El-Malki, and L. Bellver, Hierarchical Mobile IPv6 mobility management (HMIPv6), Dec IETF draft-ietfmipshop-hmipv6-04.txt. [4] R. Koodli, Fast Handovers for Mobile IPv6, Jul IETF RFC [5] ETSI, Ts , release 6, ETSI, Tech. Rep. [6] T. T. Kwon, M. Gerla, and S. Das, Mobility management for VoIP service: Mobile IP vs. SIP, IEEE Trans. Wireless Commun., vol. 9, no. 5, pp , Oct [7] Q. Wang, M. Abu-Rgheff, and A. Akram, Design and evaluation of an integrated mobile IP and SIP framework for advanced handoff management, in Proc. IEEE Int. Conf. Commun., Jun. 2004, vol. 7, pp [8] J.-W. Jung, R. Mudumbai, D. Montgomery, and H.-K. Kahng, Performance evaluation of two layered mobility management using mobile IP and session initiation protocol, in Proc. IEEE Global Telecommun. Conf., Dec. 2003, vol. 3, pp [9] N. Banerjee, K. Basu, and S. Das, Handoff delay analysis and measurement in SIP-based mobility management in wireless networks, in Proc. Int. Parallel and Distrib. Process. Symp., Apr. 2003, pp [10] W. Wu, N. Banerjee, K. Basu, and S. Das, Network assisted IP mobility support in wireless LANs, in Proc. IEEE Int. Netw. Comput. Appl., Apr. 2003, pp [11] X. Perez-Costa and M. Torrent-Moreno, A performance study of hierarchical mobile IPv6 from a system perspective, in Proc. IEEE Int. Conf. Commun., Jun. 2003, vol. 1, pp [12] A. Campbell, J. Gomez, K. Sanghyo, W. Chieh-Yih, Z. Turanyi, and A. Valko, Comparison of IP micromobility protocols, IEEE Wireless Commun., vol. 9, no. 1, pp. 2 12, Feb [13] Y. Gwon, J. Kempf, and A. Yegin, Scalability and robustness analysis of mobile IPv6, fast mobile IPv6, hierarchical mobile IPv6, and hybrid IPv6 mobility protocols using a large-scale simulation, in Proc. IEEE Int. Conf. Commun., 2004, vol. 7, pp [14] X. Costa, M. Torrent-Moreno, and H. Hartenstein, A performance comparison of mobile IPv6, hierarchical mobile IPv6, fast handovers for mobile IPv6 and their combination, Mobile Comput. Commun. Rev., vol. 7, no. 4, pp. 5 19, Oct [15] C. Perkins, Route Optimization in Mobile IP, Sep IETF Internet Draft, draft-ietf-mobileip-optim-11.txt, work in progress. Hanane Fathi (S 05 M 06) received the M.S. degree in electrical engineering from Aalborg University, Aalborg, Denmark, and the Telecommunications Engineering Diploma from Ecole Centrale d Electronique of Paris, Paris, France, both in 2002, and the Ph.D. degree in wireless communications from the Center for TeleInfrastruktur, Aalborg University in She is currently with the AIST Research Center for Information Security, Tokyo, Japan. Her research interests include VoIP over wireless networks, mobility management, authentication schemes, and wireless security. Shyam S. Chakraborty (M 04) received the M.Tech. degree from the Indian Institute of Technology, Delhi, India, and the Licenciate of Technology and Doctor of Science (Technology) degrees from Helsinki University of Technology, Espoo, Finland. He was a recipient of the Academy Fellowship from the Academy of Finland in 2000 and was a Docent with the Department of Electrical and Computer Engineering, Helsinki University of Technology. He has been a Visiting Professor with the Asian Institute of Technology, Pathumthani, Thailand, a Guest Professor with Aalborg University, Aalborg, Denmark, and a Guest Researcher with the Technical University of Berlin, Berlin, Germany. His research interests are Markov modeling of MAC and ARQ schemes, multihop ad hoc networks, diversity combining, VoIP in wireless systems, mobility management, etc. Since June 2005, he has been with Ericsson Finland, Jorvas, Finland. The work presented here was done before he joined Ericsson. Prof. Chakraborty is a Guest Editor of a special issue of the IETE Journal of Research on protocols for resource, link, and mobility management and of the IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS special issue on mesh networks. He was the General Co-Chair of the MeshNets 2005 Workshop.

11 270 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 1, JANUARY 2007 Ramjee Prasad (M 88 SM 90) was born in Babhnaur, Gaya, India, on July 1, He is now a Dutch citizen. He received the B.Sc. (Eng) degree from Bihar Institute of Technology, Sindri, India, in 1968 and the M.Sc. (Eng) and Ph.D. degrees from Birla Institute of Technology (BIT), Ranchi, India, in 1970 and 1979, respectively. Since June 1999, he has been with Aalborg University, Aalborg, Denmark, where he is currently the Director of the Center for Teleinfrastruktur (CTIF) and the Wireless Information and Multimedia Communications Chair. He is the Coordinator of the European Commission Sixth Framework Integrated Project My personal Adaptive Global NET (MAGNET). He was involved in the European ACTS Project Future Radio Wideband Multiple Access Systems (FRAMES) as a DUT Project Leader. He is a Project Leader of several international industrially funded projects. He is also an Advisor to several multinational companies. He has published more than 500 technical papers, contributed to several books, and has authored, coauthored, and edited 11 books. Dr. Prasad is a Fellow of the Institution of Electrical Engineers, a Fellow of the Institution of Electronics and Telecommunication Engineers, a member of The Netherlands Electronics and Radio Society (NERG), and a member of the Engineering Society in Denmark (IDA). He has served as a member of the Advisory and Program Committees of several IEEE international conferences. He is the Coordinating Editor and Editor-in-Chief of the Kluwer International Journal on Wireless Personal Communications and a member of the Editorial Board of other international journals, including the IEEE Communications Magazine andiee Electronics Communication Engineering Journal. He is also the Founding Chairman of the European Center of Excellence in Telecommunications, known as HERMES.