Support of Micro-Mobility in MPLS-Based Wireless Access Networks

Transcription

1 IEICE TRANS. COMMUN., VOL.E88 B, NO.7 JULY PAPER Special Section on Mobile Multimedia Communications Support of Micro-Mobility in MPLS-Based Wireless Access Networks Kaiduan XIE, Vincent W.S. WONG a), and Victor C.M. LEUNG, Nonmembers SUMMARY In this paper, we propose an architecture for the MPLS (Multi-Protocol Label Switching)-based micro-mobility management including label switched path setup, packet forwarding, handoff processing, and paging. In order to prevent packet loss during handoff, we propose two packet recovery mechanisms, namely: buffer time-based packet recovery and medium access control (MAC) layer assisted packet recovery schemes. Simulation results show that the MAC layer assisted packet recovery scheme has a better performance than buffer time-based scheme. Our proposed scheme provides a higher throughput when compared with other IP micro-mobility protocols including Cellular IP, HAWAII, and Hierarchical Mobile IP. key words: MPLS, mobility management, handoff,paging 1. Introduction Mobile IP (MIP) [1] is the current standard for supporting mobility in IP (Internet Protocol) network. It provides seamless mobility by hiding the change of IP address when a mobile host (MH) moves between IP subnets, thus providing a good framework that allows users to roam outside their home network. However, Mobile IP is not designed to support fast handoff and seamless mobility in handoff-intensive environments. Various IP micro-mobility protocols (e.g., Cellular IP [2], HAWAII [3], HFA (hierarchical foreign agent) [4]) have been proposed recently aiming to support mobility within a domain. Most of these protocols adopt a hierarchical approach by dividing the network into domains. Mobile IP is used to support mobility between two domains while intradomain mobility can be handled by micro-mobility scheme [5]. In wired networks, Multi-Protocol Label Switching (MPLS) has begun to deploy in the Internet backbone to supportservicedifferentiation and traffic engineering [6]. In MPLS, each packet is prepended with a label. MPLS simplifies the forwarding process by means of label swapping. Other advantages of MPLS include the ease of creating virtual private networks, the support of traffic engineering via constraint-based routes, and path protection via fast reroute. There has been increasing interest in applying MPLS in IP-based wireless access networks [7]. An architecture Manuscript received September 27, Manuscript revised December 21, The authors are with the Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, BC, Canada. Presently, with VectorMAX Corp. a) vincentw@ece.ubc.ca DOI: /ietcom/e88 b integrating MPLS and Mobile IP which obviates the need for IP tunneling is proposed in [8]. Label switched path (LSP) setup and handoff processing for MPLS-based micromobility management are proposed in [9]. In [10], a label edge mobility agent (LEMA) is proposed to create a hierarchical overlay network above the access network. The mechanisms for creating and modifying LSPs in MPLS and hierarchical mobile IP environments are described in [11]. Comparing with Cellular IP, HAWAII and HFA, MPLS-based micro-mobility has several advantages. It provides the capability to incorporate constraint-based routing and the support of traffic engineering in wireless access networks. The second is the ease of deployment. Within the wireless access network, only the gateway (GW), the paging server, and the base stations (BSs) need to be aware of the mobility of the MH. It improves the network reliability through the use of path protection and restoration schemes (e.g., fast reroute [6]). In this paper, an MPLS-based mobility management architecture for wireless IP networks is proposed. Our work differs from the related work ([7] [11]) in that we not only present the mechanisms for mobility management (including LSP setup, packet forwarding, handoff, paging), but we also provide qualitative comparisons with other IP micromobility management schemes. Our main contributions are as follows [12], [13]: A domain-based MPLS mobility management scheme for wireless IP networks is proposed. In order to prevent packet loss during handoff, two packet recovery schemes are proposed and evaluated. The performance of the MPLS-based mobility management scheme is studied in terms of UDP (user datagram protocol) packet loss and TCP (transmission control protocol) throughput by using ns-2 simulations. Both qualitative and quantitative performance comparisons between our proposed scheme and other IP-based mobility management schemes including Cellular IP, HAWAII, and Hierarchical Mobile IP are given. This paper is organized as follows: Sect. 2 presents our proposed architecture. In Sect. 3, we provide a qualitative comparison between our proposed scheme and Cellular IP, HAWAII, and HFA. Section 4 presents the performance comparison results via network simulation. Section 5 concludes our work and outlines the future work. Copyright c 2005 The Institute of Electronics, Information and Communication Engineers

2 2736 IEICE TRANS. COMMUN., VOL.E88 B, NO.7 JULY 2005 Fig. 1 Registration process. Fig. 2 Handoff process. 2. MPLS-Based IP Mobility Management 2.1 Label Switched Path (LSP) Setup The MIP registration protocol is used to set up an LSP for the MH. The network is divided into domains. In each domain there is an MPLS capable router acting as the gateway (GW). The GW acts as the home agent (HA) for the MHs, and handles the mobility management for the MHs in the domain. In Fig. 1, the BSs periodically broadcast the MIP advertisement (ADV) to the MHs. The identity of the BS is carried as part of the ADV. Upon receiving the ADV, the MH sends an MIP registration request (RREQ) asareply (step 1). The BS can then determine whether an MH that is powered on is in its home domain. If the MH is in its home domain, the BS sends an RREQ to the GW (step 2). The GW creates an LSP entry for the MH in its label forwarding information base (LFIB). It places the MH s address on the forwarding equivalence class (FEC) field of the LSP entry for the MH s serving BS. The GW then sends an MIP registration reply (RREP) to the BS that relays this message to the MH (steps 3 and 4). In the downlink direction, there are two LSP segments for the MH. One is from the CH s label switched router (LSR) to the GW. The other is from the GW to the MH s serving BS. In the uplink direction, all packets from the MH travel to the GW first, and are then forwarded to the destination. We assume that the LSP between the CH and the HA/GW as well as the LSP between the GW and BS are pre-configured via the label distribution protocol (LDP) [14]. In Fig. 1, all the intermediate nodes (e.g., R1, W1, GW, and BS1) are assumed to be LSRs. The MH s label information in the LFIBs of related nodes is shown in the inner table in Fig. 1. The notations are as follows: I/F: Incoming interface; I/L: Incoming Label; O/F: Outgoing interface; O/L: Outgoing Label; S: Segmented flag. The segmented flag (with a value of 1 or 0) is used to indicate the need for further LFIB lookup for packet forwarding. An active timer T a is used to decide whether or not the MH is active (i.e., transmitting or receiving data packets). Each incoming or outgoing data packet refreshes the timer T a. When the timer expires, the network assumes that the MH is in idle state, and removes the LSP entry for this user in the corresponding LFIBs. 2.2 Packet Forwarding If a CH wants to communicate with the MH, packets are transmitted via the pre-configured LSP between the CH s LSR and MH s HA. The HA knows whether there is another LSP segment for the MH via checking the segmented flag in LFIB. If the segmented flag is 1, after stripping the label and obtaining the destination address, the home agent selects an LSP using the MH s destination address as an index. All packets destined for the MH use this LSP to reach the MH s serving BS via the normal label forwarding. When packets arrive at the MH s serving BS, it relays them to the MH via wireless channel. There are three LSP segments between CH and MH: the CH s LSR to HA, HA to foreign GW, and foreign GW to the MH s serving BS. In outgoing direction, packets follow the path to the GW and are then forwarded to destination. 2.3 Handoff As in MIP, the BS periodically sends the ADV to the MH. Handoff occurs when the MH send an RREQ to the new BS (step 1 in Fig. 2). If the old and new BSs are within the same domain, the new BS informs the GW via the RREQ (step 2). The new BS also sends a handoff-notification (HFNO) to the old BS so that the old BS can forward packets to the new BS (step 3). The HFNO is a special kind of RREQ that does not need to be acknowledged. Upon receiving RREQ, the GW updates the MH s entry in its LFIB. It places the

3 XIE et al.: SUPPORT OF MICRO-MOBILITY IN MPLS-BASED WIRELESS ACCESS NETWORKS 2737 MH s address in the FEC column of the entry for the new BS, and sends an RREP (step 4). The new BS then sends an RREP message to MH (step 5). Upon receiving HFNOfrom the new BS, the old BS adds an LSP entry in its LFIB for the MH by including the MH s address in the FEC column of the LSP entry for the new BS and starts the active timer T a. At the same time, the old BS sets the segmented flag to 1 in the entry for the LSP from GW to itself. From this point on, the old BS forwards packets to the new BS on the appropriate LSP until the timer T a expires. 2.4 Paging For every handoff the MH sends an RREQ to the GW no matter how close the two neighboring BSs are. This is not effective when the covering area of the GW is large. To overcome this shortcoming, an extra level of hierarchy is introduced. In addition, in order to reduce the power consumption of MHs and the signalling load in the access network (caused by frequent registrations and handoffs), we propose to employ paging. The wireless access network domain is divided into different paging areas. We assume that there is a paging server (PS) in each paging area. After introducing PS, the GW only maintains the location information of that MH in the granularity of a paging area. The identity of the PS is carried as part of the ADV from BS. When an idle MH crosses a paging boundary, it sends an RREQ to the GW to report its current paging area. When an idle MH crosses a cell boundary, but stays within the same paging area, it does not need to send an RREQ to the PS to report its location. When an MH is in its active state, upon every cell boundary crossing, the MH sends an RREQ to PS. The PS knows whether or not the MH has moved to another PS from this RREQ. If this movement spans two PSs, the PS sends an RREQ to the GW to report this handoff. The GW updates the routing entry for MH in its LFIB. When an MH is in its active state, in the forward path (from GW to MH), incoming packets destined for MH are intercepted by GW. The GW then forwards those packets to the corresponding PS. The PS forwards those packets to the MH s serving BS. In the reverse path (from MH to GW), outgoing packets are sent to PS first. The PS then forwards those packets to GW. The GW forwards those packets to the appropriate destinations. When an MH is in its idle state, the PS buffers all incoming packets corresponding to that MH and sends a paging request (PREQ) to all BSs within its paging area via multicast (step 1 in Fig. 3). Those BSs then send a layer 2 (link layer) PREQ via wireless channel (step 2). Upon receiving the PREQ, the MH sends an RREQ message to BS (step 3). The BS sends an RREQ message to the PS (step 4). The PS installs an LSP entry for MH and forwards buffered packets. 2.5 Packet Loss Recovery In Sect. 2.3, when a handoff occurs, the new BS informs the Fig. 3 Paging process. old BS so that the old BS can set up an LSP for MH. Subsequent packets follow this path to the new BS that relays them to the MH. However, before the old BS receives the HFNO message from the new BS, the packets destined for MH still take the path to the MH via the old BS. Unfortunately, the MH has moved to the new BS, so these packets are eventually lost on wireless channel. Depending on the network topology and the wireless network capability, these dropped packets due to handoff may decrease the application s performance. To minimize packet loss, two alternate methods may be used as described below Buffer Time-Based Packet Recovery The use of buffer timer was originally proposed in [2] for Cellular IP. In this paper, we extend the idea to work in MPLS environment. During handoff, all the incoming packets destined for the MH are buffered at the old BS. The old BS also stamps each packet the instant the packet arrives. After receiving the HFNO from the new BS, the old BS uses the buffering time T b to decide which packets should be forwarded to the new BS. If the elapsed time between the arrival instant and the current time is smaller or equal to T b, the packet is forwarded to the new BS. Otherwise, it is discarded. The optimal T b causing zero packet loss is equal to the sum of the second layer handoff time and the link delay between the old and new BSs. Since it is difficult to estimate the optimal T b for each handoff, a sub-optimal value is used. We assume that T b is equal to 0.5 T beacon where T beacon denotes the interval at which the BS emits the beacon signal to MHs MAC Layer Assisted Packet Recovery In wireless LANs [15], the MAC layer uses the Request-To- Send (RTS), Clear-To-Send (CTS), DATA, and Acknowledgment (ACK) for data exchange. The MAC layer uses the RTS control frame and a short CTS frame to reserve channel

4 2738 IEICE TRANS. COMMUN., VOL.E88 B, NO.7 JULY 2005 Fig. 4 MAC layer assisted packet recovery. access. After data transmission, the ACK frame confirms the success of the transmission. Our proposed MAC layer assisted packet recovery scheme is designed to prevent the packet loss due to handoff. The packet loss during handoff is caused by the broken or deteriorated wireless connection between the old BS and the MH. In our proposed scheme, a buffer in the MAC layer is used to cache the packets dropped by the MAC layer. These packets are then transmitted from the old BS to the new BS. In Fig. 4(a), IP packets are passed from network layer to link layer (LL). A MAC layer frame is constructed within the LL module. The MAC frame is then placed in the interface queue (IFq) for transmission. The MAC module in Fig. 4(a) is responsible for channel access. If a particular frame transmission is not successful after several attempts, this frame is then placed in the MAC buffer. After receiving a HFNO from new BS, the old BS forwards those frames in the MAC buffer destined to that MH to the new BS via the pre-configured LSP. After that, those frames in the IFq buffer destined to that MH are also forwarded to the new BS. All MHs within the coverage of the same BS share the MAC buffer. The BS uses the MH s MAC address carried in the HFNO to identify which packets belong to the corresponding MH. In this way, we can eliminate the packet loss due to handoff and thus improve the application performance. The MAC layer assisted packet recovery process is illustrated in Fig. 4(b). 3. Qualitative Comparisons We compare several IP mobility management schemes. We call the schemes proposed in [9] and [10] the basic MPLS mobility management (BMMM) and LEMA, respectively. Our scheme is referred to as enhanced MPLS mobility management (EMMM) [12], [13]. Registration: MIP, or its extension is used in EMMM, HFA and HAWAII, while Cellular IP uses its own proprietary protocol. There is no specific information with regard to whether or not MIP is used in BMMM and LEMA. Location Information Create or Refresh: In EMMM, LSP is setup using control packets, but refreshed using data packets originating from or destined to MH. In HAWAII and HFA, location information for MH is created and refreshed by control packets. In Cellular IP, location information for MH is created by data packets and refreshed by data or control packets. No information is given in [9], [10] regarding the mechanism to refresh the location informationinbmmmandlema. Paging Support: In EMMM and HAWAII, explicit paging is adopted. When the MH is in its idle state, a paging packet is sent by the related node (the PS in EMMM, the selected node in HAWAII). However, in Cellular IP, paging is performed in an implicit way. The paging cache is used to route the packet. In basic HFA, no paging is supported. There is a proposal to introduce paging in MIP [16], which also adopts explicit paging. There is no information regarding paging in BMMM and LEMA. QoS Support: EMMM relies on MPLS for QoS support, such as MPLS-based Diff-Serv and Inter-Serv. HFA and HAWAII can provide IP-based QoS since they work on classical IP routing. However, there is no inherent support for QoS in Cellular IP. BMMM and LEMA can also provide QoS support based on the capability of MPLS. Traffic Engineering: EMMM can use constraint based routing to provide traffic engineering. HFA and HAWAII cannot support traffic engineering since conventional IP routing cannot provide traffic engineering. In Cellular IP, supporting traffic engineering is more difficult. BMMMand LEMA can also provide traffic engineering since this is the capability of MPLS. Reliability: Constraint based routing based fast reroute can be used to protect link or node failure in EMMM. HFA and HAWAII rely on the underlying IP capabilities, for example, the OSPF (Open Shortest Path First) protocol to distribute the link information and re-compute the path, which is slower than fast-reroute. In Cellular IP, no special schemes are used. The network has to rely on the periodic beacon signal transmitted by the GW and the routing refresh sent by the MH. BMMM and LEMA have the same capability as our proposed scheme. Packet Recovery: MAC assisted packet recovery is proposed to eliminate packet loss due to handoff in EMMM. In Cellular IP, semi-soft handoff with delay device is proposed to cope with packet loss due to handoff. However, it is difficult to estimate the accurate value of the delay. In HAWAII, UNF (unicast non-forwarding) can utilize the MH s capability to listen to the neighboring BSs simultaneously to provide better service. In HFA, BMMM and LEMA, no special methods are described to reduce packet loss. Deployment Requirements: Only the BSs, GW, and PSs need to be retrofitted in EMMM to support mobility management, provided that all nodes in the domain are MPLS-enabled IP routers. In HFA, only the GW FA and

5 XIE et al.: SUPPORT OF MICRO-MOBILITY IN MPLS-BASED WIRELESS ACCESS NETWORKS 2739 regional FA need to be retrofitted, while in HAWAII and Cellular IP, all nodes in the domain need to be retrofitted to support mobility management. In BMMM, only the GW needs to be upgraded. For LEMA, some nodes in the access network should be upgraded to configure LEMA capable software. However, it is difficult to implement customized LEMA chain for each MH based on the mobility pattern of the MH, the available bandwidth and other factors. Further discussion can be found in [13]. 4. Performance Analysis We use the ns-2 version 2.1b6 [17] as our simulation tool with MPLS module [18] and IP micro-mobility module CIMS (Columbia IP Micro-Mobility Software) [19]. We extend the MPLS module to hierarchical address format as needed for MIP based simulation. The network topology is shown in Fig. 5, which is assumed to be the MH s home network. LSRs W1 and W2 are the PSs. In the network each wired connection is modeled as a 10 Mb/s duplex link with link delay of 10 ms. Since the BSs are assumed to be LSRs with mobility management capabilities, we assume that there is a direct connection between each pair of neighboring BS. The MH connects to the BS using ns-2 IEEE model. The coordinates for four BSs are (0, 0), (140, 140), (280, 280), (420, 420) where the unit is meters. The BS broadcasts an ADV message at an interval of 0.5 s. In order to model the scenarios where neighboring BSs can overlap with each other, we adjust the transmitting power of BSs according to the overlapping distance, denoted by d(overlap). Figure 6 shows the concept of overlapping distance. When the MH crosses the cell boundary, upon receiving the first beacon signal from the new BS, the MH assumes that a handoff has occurred and notifies the BS [20]. tance is 0 m. We use RTP sequence number to record UDP packets. In Fig. 7(a), the UDP packet number is shown during the first handoff when the MH moves from BS1 to BS2. No buffer time-based packet recovery is applied. We can observethepacketlossduringthehandoff period. Figures 7(b) and (c) show the packet number when the MH moves from BS1 to BS2, and then from BS2 to BS3, respectively. Buffer time-based packet recovery scheme is used with the buffer time T b of 250 ms. The buffer time is approximately half the ADV interval, plus the delay between the new and old BS. Figure 7(b) shows that when T b is larger than the actual handoff time, packet duplications may occur. On the other hand, Fig. 7(c) shows that when T b is smaller than the actual handoff time, a small number of packets may be lost TCP Performance The MH moves from BS1 to BS2, and then from BS2 to BS3. In Fig. 8(a), buffer time-based recovery is not used. Due to packet loss, TCP slow start is initiated [21]. In Figs. 8(b) and (c), buffer time-based packet recovery is used with T b of 250 ms. Figure 8(b) shows that when T b is larger than the actual handoff time, packet duplications may occur. Figure 8(c) shows that when T b is smaller than the actual handoff time, a small number of packets may be lost, triggering the TCP slow start. Figure 9 shows the throughput for TCP new Reno with and without buffer time-based packet 4.1 Buffer Time-Based Packet Recovery UDP Packet Loss In this experiment, UDP probing traffic is directed from the CH to the MH and consists of 210 bytes packets transmitted at 10 ms interval. The MH moves back and forth between BS1 and BS4 with a speed of 20 m/s. The overlapping dis- Fig. 6 Illustration of overlapping distance. Fig. 5 Simulation network topology. Fig. 7 UDP with buffer time-based packet recovery.

6 2740 IEICE TRANS. COMMUN., VOL.E88 B, NO.7 JULY 2005 Fig. 8 Packet number with buffer time-based packet recovery. Fig. 10 UDP packet number without packet recovery. Fig. 9 Throughput with buffer time-based packet recovery. Fig. 11 UDP with MAC layer assisted packet recovery. recovery. T b is equal to 250 ms. Although buffer time-based packet recovery in the old BS can improve the TCP throughput, it is difficult to estimate the accurate buffer time value for each handoff. 4.2 MAC Layer Assisted Packet Recovery UDP Packet Loss UDP probing traffic isdirected from CHto MHandconsists of 210 bytes packets transmitted at 10 ms interval. The MH moves from BS1 to BS2 at a speed of 20 m/s. Figure 10 shows the packet loss caused by handoff without packet recovery scheme under different overlapping distances. The numbers of lost packets are 22, 9, 0 with overlapping distance of 0, 5 and 10 meters, respectively. The time during which packets are lost is equal to the sum of data link layer handoff time and the delay between the new BS and PS. Figure 10 shows that overlapping BSs can reduce packet loss during handoff. However, in some networks (e.g., GSM), the MH cannot listen to two neighboring BSs simultaneously. Therefore, it is still necessary to minimize the packet loss due to handoff. Figure 11 shows that the MAC layer assisted packet recovery can totally eliminate packet loss during handoff. As the overlapping distance increases, the time gap between the packet stream from the old BS to the MH, and the other stream from the old BS via the new BS to the MH becomes smaller. Comparing Figs. 11(b) and (c), we can see that when the overlapping distance is large enough, the MAC layer assisted packet recovery has no gain because the MH can connect to two neighboring BSs simultaneously for a time that is long enough to accommodate the handoff TCP with MAC Layer Assisted Packet Recovery In Fig. 5, the MH moves from BS1 to BS2 at a speed of

7 XIE et al.: SUPPORT OF MICRO-MOBILITY IN MPLS-BASED WIRELESS ACCESS NETWORKS 2741 Fig. 12 TCP with MAC layer assisted packet recovery. Fig. 13 Different overlapping distance. 20 m/s, the overlapping distance is 0 meter. Figure 12(a) shows the TCP Reno behavior during handoff without packet recovery. At time s, the MH receives the beacon signal from the new BS. TCP slow start is initiated due to timeout. Figure 12(b) shows the TCP packet number observed at the MH with the use of MAC layer assisted packet recovery. No TCP packet is either dropped or retransmitted. The gap between the two packets streams (210 ms in Fig. 12(b)) is equal to the handoff time plus the link delay from new BS to old BS via which the HFNO message traverses TCP with Different Overlapping Distance In Fig. 5, the MH moves back and forth between BS1 and BS4 at various speeds. The simulation time is 400 s. Figure 13 shows that a higher TCP throughput can be achieved with the use of the MAC layer assisted packet recovery. The larger the overlapping distance between neighboring BSs, the higher the throughput. With a larger overlapping distance, the MH can receive beacon signal from the new BS sooner and can also maintain the connection with the old BS longer. The throughput decreases as the speed of the MH increases Comparisons with IP Micro-Mobility Protocols We compare between our proposed scheme with Cellular IP hard and semi-soft handoff [2], HAWAII UNF (unicast non-forwarding) and MSF (multiple stream forwarding) [3], and HFA [4]. In order to provide a fair comparison, we made some modifications in HFA and HAWAII module from CIMS. In the original CIMS, upon receiving the first beacon signal from the new BS, the connection between the MH and the old BS is cut off to simulate the network where MH can only listen one BS s signal at any given time. We model this by adjusting the transmitting power of the BS. In Cellular IP: the route update interval value is 0.5 s; the rout- Fig. 14 TCP throughput with overlapping distance of 0 meter. ing cache time out value is 1.5 s; the paging cache time out value is 15 s; the active time value is 2 s; and the semi-soft delay is 50 ms. In Fig. 5, all nodes except CH, R1 and R2 are assumed to be Cellular IP capable nodes and are configured with routing cache and paging cache. In HAWAII and HFA, beacon signal is transmitted at an interval of 0.5 s. Figure 14 shows that our proposed scheme with MAC layer assisted packet recovery outperforms all other IP micro-mobility schemes. The performance between our proposed scheme without MAC layer assisted packet recovery and HFA is very close since HFA does not use any packet recovery scheme. HAWAII MSF gains a small edge over HFA, MPLS without packet recovery as well as Cellular IP semi-soft handoff. When no packet recovery scheme is used, the performance of MSF and UNF in HAWAII is very close. The advantage of MSF over UNF is that packet recovery can be applied in the old BS. Note that in MSF, packets are first forwarded from the old BS to the new BS before they are diverted at the crossover router. In addition, we observe that in Cellular IP, periodical route-update and paging-update messages compete over the shared wireless

8 2742 IEICE TRANS. COMMUN., VOL.E88 B, NO.7 JULY 2005 channel with the data packets. Cellular IP gives a lower TCP throughput when compared with other schemes. We also perform simulations with overlapping distance of 5 m and 10 m [13]. The results are similar to that with overlapping distance of 0 m. 5. Conclusions We have proposed an architecture for MPLS-based micromobility management in wireless access networks. We have presented the procedures for LSP setup, packets forwarding, handoff, and paging. To reduce packet loss due to handoff, buffer time-based packet recovery and MAC layer assisted packet recovery schemes are proposed. Simulation results indicated that both packet recovery schemes can effectively prevent packet loss during handoff. The MAC layer assisted scheme is preferred in practice since it may not be simple to estimate the buffer time value for the buffer time-based packet recovery scheme. On TCP throughput, the MPLSbased micro-mobility with MAC layer assisted packet recovery outperforms other IP micro-mobility schemes (Cellular IP, HAWAII, Hierarchical MIP). Our work shows that MPLS is feasible to support mobility management in IPbased wireless access networks. Further work includes the study of traffic engineering and admission control in MPLSbased wireless IP network, as well as internetworking with the all IP-core network domain [22]. References [1] C. Perkins ed., IP mobility support for IPv4, IETF RFC 3344, Aug [2] A. Campbell, J. Gomez, S. Kim, A. Valko, C.Y. Wan, and Z. Turanyi, Design, implementation and evaluation of cellular IP, IEEE Pers. Commun., vol.7, no.4, pp.42 49, Aug [3] R. Ramjee, K. Varadhan, L. Salgarelli, S.R. Thuel, S.-Y. Wang, and T. La Porta, HAWAII: A domain-based approach for supporting mobility in wide-area wireless networks, IEEE/ACM Trans. Netw., vol.10, pp , June [4] E. Gustafsson, A. Jonsson, and C. Perkins, Mobile IPv4 regional registration, IETF Internet Draft, draft-ietf-mobileip-reg-tunnel- 07.txt, Oct [5] A.T. Campbell, J. Gomez, S. Kim, C.-Y. Wan, Z.R. Turanyi, and A.G. Valko, Comparison of IP micro-mobility protocols, IEEE Wireless Commun. Mag., vol.9, no.1, pp.72 82, Feb [6] E. Rosen, A. Viswanathan, and R. Callon, Multiprotocol label switching architecture, IETF RFC 3031, Jan [7] Y. Guo, Z. Antoniou, and S. Dixit, IP transport in 3G radio access networks: An MPLS-based approach, Proc. IEEE WCNC 02, pp.11 17, March [8] Z. Ren, C.-K. Tham, C.-C. Foo, and C.-C. Ko, Integration of mobile IP and multi-protocol label switching, Proc. IEEE ICC 01, pp , June [9] H. Kim, K.S.D. Wong, W. Chen, and C.L. Lau, Mobility-aware MPLS in IP-based wireless access networks, Proc. IEEE Globecom 01, pp , Nov [10] F.M. Chiussi, D.A. Khotimsky, and S. Krishnan, A network architecture for MPLS-based micro-mobility, Proc. IEEE WCNC 02, pp , March [11] V. Vassiliou, H.L. Owen, D.A. Barlow, J. Grimminger, H.P. Huth, and J. Sokol, A radio access network for next generation wireless networks based on multi-protocol label switching and hierarchical mobile IP, Proc. IEEE VTC-Fall, pp , Sept [12] K.D. Xie, V.W.S. Wong, and V.C.M. Leung, Support of micromobility in MPLS-based wireless access networks, Proc. IEEE WCNC 03, pp , March [13] K.D. Xie, MPLS-based Mobility Management for Wireless Cellular Networks, Master s thesis, University of British Columbia, Canada, April [14] L. Andersson, P. Doolan, N. Feldman, A. Fredette, and B. Thomas, LDP specification, IETF RFC 3036, Jan [15] IEEE standard, [16] X. Zhang, J.G. Castellanos, and A.T. Campbell, P-MIP: Paging extensions fot mobile IP, ACM Mobile Networks and Applications, vol.7, no.2, pp , April [17] [18] fog1/mns/ [19] [20] C. Perkins, Mobile IP Design Principles and Practices, Addison- Wesley Longman, Reading, [21] W.R. Stevens, TCP/IP Illustrated: The Protocols, Addison Wesley Professional, [22] 3GPP2 Specification, All IP-Core Network Multimedia Domain, version 1.0, TSG-X.S , Dec Kaiduan Xie received the bachelor degree from Hubei University in 1993, Master of Engineering degree from Beijing University of Posts and Telecommunications in 1996, and M.A.Sc. degree from the University of British Columbia in Currently, he is a senior software engineer at VectorMax.com Corp., New York, USA. Vincent W.S. Wong received the B.Sc. degree from the University of Manitoba, in 1994, the M.A.Sc. degree from the University of Waterloo, in 1996, and the Ph.D. degree from the University of British Columbia (UBC), in He is currently an assistant professor in the Department of Electrical and Computer Engineering at UBC. His research interests are in wireless communications and networking. Dr. Wong received the Natural Science and Engineering Research Council postgraduate scholarship and the Fessenden Postgraduate Scholarship from Communications Research Centre, Industry Canada, during his graduate studies. Victor C.M. Leung received the B.A.Sc. and Ph.D. degrees from the University of British Columbia (UBC) in 1977 and 1981, respectively, where he was awarded the APEBC Gold Medal and the NSERC Postgraduate Scholarship. After working in MPR Teltech Ltd. and the Chinese University of Hong Kong, he returned to UBC in 1989, where he is a Professor of Electrical and Computer Engineering and TELUS Mobility Research Chair in Advanced Telecommunications Engineering. His research interests are in protocols and management of computer and telecommunication networks. Dr. Leung is a Fellow of IEEE, a voting member of ACM, and an editor of both the IEEE Transactions on Wireless Communications and Vehicular Technology.