Efficient Mobility Management for Seamless Roaming between WWAN and WLAN

Transcription

1 Efficient Mobility Management for Seamless Roaming between WWAN and WLAN Qian Zhang, Chuanxiong Guo, Zihua Guo, and Wenwu Zhu Wireless and Networking Group Microsoft Research Asia, No. 49, Zhichun Road, Haidian District, Beijing, China, ABSTRACT As we are moving toward the next generation all-ip wireless networks, we are facing the integration of non-homogeneous networks, such as wireless wide area networks (WWAN) and wireless local area networks (WLAN). In vertical handoff between WWAN and WLAN, the mobile hosts are required to be able to move freely across different networks while satisfying the quality of service requirements for a variety of applications. In order to maintain continuity of connection and seamless roaming with particular emphasis on the data applications, in this paper, we propose novel mobility management system which integrates a connection manager (CM) that can intelligently detect the network conditions change and a virtual connectivity manager (VC) that uses end-to-end principle to maintain the session connectivity. In our solution, handoff rate and ping-pong effect can be significantly reduced using CM. Meanwhile, mobility can be made transparent to application without network infrastructure support using VC, which can also handle mobility under Network Address Translation (NAT) and simultaneous movements scenarios. Experiment results demonstrate the effectiveness of our proposed schemes. Our prototyping system shows that seamless roaming between WLAN and WWAN can be achieved while much better performance can be obtained comparing with the one using traditional scheme. 1. INTRODUCTION The increasing amount of Internet users in combination with the evolution of IP-based applications has created a strong demand for wide-area, broadband access to IP services. A future wireless Internet is expected to consist of different types of wireless networks each of which provides varying access speed and coverage level. Wireless LANs can complement with the next-generation cellular networks by offering a cost-efficient, wireless broadband data solution for hot spot areas. Today, the natural trend has been towards utilizing high-bandwidth data networks such as IEEE in hot spots whenever they are available and switching to an overlay service such as GPRS/UMTS networks when the coverage of WLAN is not available or the network condition in WLAN is not good enough. We refer to such a procedure as inter-network roaming or vertical handoff. By combining the wide coverage of next-

2 generation cellular systems with the advantages of speed and capacity of wireless LANs, users can make the most of wireless IP communication. As a result, in next-generation wireless systems, one of the major features that differs from the current communication service systems is seamless and efficient global roaming. This requires that the mobile subscribers can move freely across different networks while maintaining their quality of service (QoS) for a variety of applications. Notice that in here there are two directional movements, one is from small coverage network (e.g., WLAN) to larger coverage network (e.g., WWAN), and the other is the reverse direction. In order to achieve seamless roaming, several issues, such as handoff metrics and handoff decision algorithms, mobility handling to maintain the on-going user connections, need to be addressed. While significant work has been done on handoff mechanisms in circuit switched mobile networks, there is not much work in literature available for packet switched mobile networks. Existing performance measures such as call blocking, call dropping etc., are applicable only to real time traffic and may not be suitable for bursty data traffic, such as file transferring and Web browsing. To date the handoff schemes that only deal with the switch between the base stations (BSs) or access points (APs) which in the homogeneous wireless system had been well studied [1]. They are usually called horizontal handoff. However, in the case of vertical handoff, we are faced the following challenges: When user moving from WWAN to WLAN, since the WWAN is usually always on, the handoff can not be triggered by the signal decay of current system as in horizontal handoff [2]. In vertical handoff between WWAN and WLAN, there is no comparable signal strength available to aid the decision as in horizontal handoff. Therefore, we need to design an efficient scheme to timely detect the WLAN availability/unavailability while reducing the ping-pong effect and handoff rate. During handoff procedure, the metrics the applications really interested in are network conditions, such as available bandwidth and delay, user preference, etc., rather than the physical layer parameters such as received signal strength (RSS), path loss, signal-to-interference ratio (SIR), etc. Thus, one should be able to perform the upper layer sensing in addition to the physical layer sensing to obtain the wireless network conditions so that the handoff is beneficial to the ongoing service in terms of QoS. Some work for the vertical handoff has been reported in literatures. In [3], a roaming scheme that considered the relative bandwidth of WLAN and GPRS was proposed. However, the authors did not provide the scheme on how to obtain the bandwidth. In [4], a detailed vertical handoff signaling procedure was presented. However, no details on the handoff decision algorithm were considered. The core of a roaming system is the handoff/mobility management scheme, which can maintain user connections after a vertical handoff. Mobile IP is the most widely studied approach for handling mobility,

3 where packets from and to the mobile node are tunneled through a home agent at its home network so that the corresponding node that communicates with the mobile host can be shielded from the mobility of the mobile node [5] [6]. To improve the routing performance and resolve certain scalability problem associated with mobile IP, several new schemes have been proposed [7]. All of these solutions, including Mobile IP, significantly rely on the network infrastructure such as home agent. Moreover, Mobile IP is not sufficient for real-time connections because it incurs high jitter, long latency and disruptive handoffs. Session Initiation Protocol (SIP) extensions have been proposed to extend the protocol in order to support terminal mobility [8], alleviating some of the shortcomings associated with Mobile IP and its route optimization variants. However, this approach applies only for real-time communications over UDP. Target at reducing the significant cost, complexity, and performance degradation coming from Mobile IP solution, and most importantly allowing for an easier deployment path than Mobile IP, an end-to-end solution named Migrate [9] has been proposed. Although it has many attractive characteristics, it has two major limitations as follows. 1) Since a node notifies the IP address change directly to its peer, this scheme cannot make mobility transparent to applications. That is to say, the legacy UDP-based applications need to be re-coded to support mobility; 2) As a pure end-to-end approach, it can not maintain user connections under several cases such as when both nodes move simultaneously and when nodes are behind Network Address Translation (NAT) box. To provide an efficient seamless roaming service to end users, we propose a completely IP-centric approach to address all the above mentioned issues. More specifically, a connection manager (CM) is introduced to intelligently detect the conditions of the different types of networks and the availability of multiple networks. CM can handle two directional roaming between WWAN and WLAN: from WWAN to WLAN and from WLAN to WWAN. When a user in a WWAN system moves into a WLAN system, since the WWAN connection is always on, the proactive roaming procedure is performed. Both the physical layer and the MAC layer sensing of WLAN system is performed to assure better quality of service and less cost while roaming from WWAN to WLAN. On the other hand, when the user moves out of the WLAN area, we should timely detect the unavailability of the WLAN and switch the wireless connection from WLAN to WWAN seamlessly. Therein an FFT-based signal decay detection scheme is used to reduce the ping-pong effect, and an adaptive threshold configuration approach is proposed to prolong the time that user stays in WLAN. Meanwhile, a virtual connectivity manager (VC) is proposed to maintain the user sessions by using an end-to-end argument when roaming occurs. By utilizing the information provided by CM, VC can not only maintain the connection unbroken, but also always achieve the best possible communication quality. In VC, in order to address the aforementioned problems faced by end-to-end schemes, 1) we introduce a local connection translation (LCT), which maintains a mapping relationship between the original

4 connection information and the current connection information for each active connection, making this mobility solution transparent to upper applications; 2) a subscribe/notify (S/N) service, which provides a bridge between two communicating parties, is proposed to support the NAT and simultaneous movements cases. The collaboration between CM and VC accomplishes seamless and efficient roaming. 2. ARCHITECTURE OF SEAMLESS ROAMING BETWEEN WWAN AND WLAN In order to provide seamless services to users while they are moving across the different networks, we focus on a completely IP-centric approach that is anticipated to further drive the notion of constituting IP, the dominant internetworking technology of the future. Motivated by this, we propose an architecture to achieve seamless roaming as illustrated in Figure 1. Figure 1. Architecture for seamless roaming. The key components of this architecture consist of connection manager (CM), virtual connectivity manger (VC), roaming decision maker, and context. Among them, roaming decision maker and context are acting as the interconnection between CM and VC. When user moves across the different wireless networks, CM will monitor the network conditions, availability of the networks, and the final network change. Based on certain measurement mechanism, CM can determine whether a roaming is needed. Corresponding signaling is sent to network context and then forwarded to VC to notify the roaming event, and leave the decision right to the roaming decision maker (based on the information from upper application/service layer). We break through the traditional vertical handoff model and introduce novel proactive roaming concept, which means users/applications involve in the roaming decision and have the final right for the decision. Users /applications involvement is based on the fact that different upper layer services or applications have different characteristics and different requirements, e.g., some service may have higher security requirements, etc.

5 When mobile user roams across different networks, VC adopts an end-to-end approach to handle the mobile routing issue for IP connectivity. VC resides between IP layer and transport layer, thus it works for both IPv4 and IPv6 in a natural way. No network infrastructure support is needed in our scheme. Triggered by the signal from roaming decision maker, which is eventually sent from CM, VC maintains the connection that can achieve the best possible quality. By introducing the local connection translation concept, VC makes mobility transparent to the upper layer applications. By introducing the S/N service, the situation when mobile node is under NAT and both of the mobile nodes are moving simultaneously can be handled efficiently. Context-awareness is considered as a fundamental property of future personalized mobile applications, where services not only need to adapt to the user s needs and preferences, but also must be aware of the device and network characteristics. The context in this sense consists of the following aspects: 1) user aspect, such as the needs, preferences, history, and behavior of the user; 2) location-related aspects, such as physical coordinates, velocity, and ambient conditions; 3) technical aspects, such as bandwidth of the network and capabilities of the terminal, etc. The final roaming decision is made take the entire related context into consideration. In the next two sections, we will describe the CM and VC module in detail. 3. CONNECTION MANGER The objective of the CM is to intelligently sense the network change and provide the seamless and proactive roaming collaborated with VC. In our work, we consider two roaming scenarios. When moving from WWAN to WLAN, since WLAN is optional, the objective of the roaming is to improve the QoS. When moving out side of WLAN, we need to have a timely and accurate handoff decision to maintain the connectivity before the loss of WLAN link. 3.1 Roaming from WWAN to WLAN As we mentioned above, when a user who is connected to a WWAN system steps into a WLAN area, Since the WWAN connection is always on, the user would like to change the connection to WLAN to obtain possible larger bandwidth and less cost. That is to say, the goal of the roaming into WLAN is for better QoS. To assure this, we need to perform the sensing in both the physical layer and the MAC layer of WLAN system. More specifically, physical layer sensing is used to detect the availability of the stable WLAN signal. MAC layer sensing is used to detect the network conditions of the WLAN system, for example, the access delay and the available bandwidth. We will present the physical-layer sensing and MAC-layer sensing next Physical Layer Sensing

6 Since CM is built above the hardware driver, it is very convenient to simply rely on the hardware itself to detect the availability of WLAN. Note that after finding a stable WLAN signal, the card itself will set some indications in the interface with upper layers to indicate the availability of the WLAN. For example, in Windows, with the Network Driver Interface Specification (NDIS) interface, we can obtain some valid objects values as long as the WLAN signal is ready, even though the device hasn t really joined the WLAN system, e.g., the object OID_802_11_BSSID_LIST or OID_802_11_BSSID. Additionally, in order to reduce the ping-pong effect around the boundary, we require that the WLAN signal should be stable enough. That is, the mean received signal strength index (RSSI) should larger than S 1 for some time, where S 1 is a threshold that will be adaptively set, as described later MAC Layer Sensing MAC layer sensing is used to estimate the network conditions of WLAN system, e.g., the available bandwidth or the access delay, before joining this WLAN system. In this work, we propose to listen and collect the Network Allocation Vector (NAV) in the MAC layer to estimate the network conditions. It is known that NAV is the major mechanism in the MAC of IEEE WLAN system to avoid collision. Once a station hears the other stations transmission, it will set the NAV to be busy state and keep silent for the time duration equals to the duration ID in the packet header [10]. From above description, we can see that the NAV busy occupation can well reflect the channel s busy state or traffic load. The higher the traffic, the larger the NAV busy occupation will be, and vice versa. With a lot of simulations and analysis, we have found that there is a fixed relationship between NAV and the available bandwidth and the latency, which can be illustrated as Figure 2(a) and Figure 2(b), respectively. This relationship is insensitive to the user number and the traffic pattern. That is, once we observe a NAV value in a time window, the residual bandwidth and latency can be estimated given certain packet length. An alternative approach to do the MAC sensing is proactively sending some probing packets. From the probing results, the collision probability can be calculated which can be mapped to the available bandwidth and delay. Nevertheless, due to the lack of space, we will not show details here. 2.5 Residual BW (Mbps) Frame size = 4 kbits Frame size = 8 kbits NAV Occupation (a) Residual /Available Bandwidth.

7 Delay (ms) NAV Occupation (b) Mean delay. Figure 2. Network performance under different NAV occupation. In summary, the MAC sensing scheme has several characteristics. First, a network condition-aware roaming is achieved. Second, the QoS information can facilitate the adaptation of upper network and application layers. Thirdly, among multiple APs, the one with the best QoS can be selected. 3.2 Roaming from WLAN to WWAN Since WWAN has a smaller coverage, when the user steps out of the WLAN area, we should timely detect the unavailability of the WLAN and switch the connection to WWAN seamlessly. Therefore, the goal for this directional roaming is to switch to WWAN before the WLAN link breaks, and meanwhile stay in WLAN as long as possible due to less cost and better QoS. In general, there are two key issues in detecting the WLAN unavailability. How to accurately detect the signal decay? Although in average, the mean RSSI will decrease when user leaves, there is quite large variation in the sampled RSSI (e.g., up to 10 db). How to determine the signal is weak? Note that different OEM has different card implementation technique although the standard defines a lower bound for WLAN signal. To handle these two problems, an FFT based approach is proposed to detect the signal decay. Meanwhile, an adaptive configuration scheme is proposed to set appropriate RSSI threshold FFT Based Decay Detection To detect the signal decay, we propose an FFT based decay detection approach by referring to the following lemma, which is easy to prove. Lemma: The fundamental term of the FFT of a statistically decreasing sequence with length N always has a negative imaginary part. That is, N 1 2 (1) = πn E X x( n) sin( ) < 0. n= 0 N The advantages of using the FFT to detect the signal decay are summarized as follows.

8 The FFT fundamental term is much more sensitive than the normal x method since the first N/2 signals should have a quite large difference between the second N/2 signals even with variation. 2πn If we regard sin( ) as a linear filter applied to the sequence x(n), it is obvious that X(1) is the N 2πn most smooth metric because sin( ) is the filter with the least high frequency component. This N will reduce the variation of X(1) even x(n) may vary severely. Based on the above lemma, we can set a threshold for X(1)/N. If the result is smaller than this threshold, the signal should be decaying Adaptive Threshold Configuration As different OEM has different implementation technique, different WLAN NIC card may have different performance. Therefore, it is necessary to set the threshold for weak signal adaptively according to each card s performance. In this work, we apply a max-min method to approach this card performance, which is described as follows. Step1. When sampling the RSSI, record current RSSI if the value of object OID_802_11_BSSID is valid. Step2. If the sampled RSSI<S 2 for some duration (say, 1 second), then update S 2 with the maximum RSSI within this time duration. Correspondingly, set S 1 =S 2 +, where is a margin. Figure 3. The diagram for CM when roaming from WLAN to WWAN. The whole vertical handoff procedure in the connection manager for WLAN to WWAN is depicted in Figure 3, which is a two-stage procedure. When the client is in WLAN, the RSSI will be sampled periodically. Once the RSSI is less than S 1, an intensive sampling will begin. If the RSSI is consistently less than S 1 for a period of time or the average RSSI is less than S 1, then we perform the FFT and compare with the FFT threshold. If the imaginary part of the fundamental term of the FFT is small enough, that

9 means the currently associated AP is weak and is to loss connection. Then, we will check if there are other strong APs nearby. If yes, we will simply wait for the intra-wlan handoff. Otherwise, we need to pass the WLAN_WEAK message to upper layers. After the upper layers actions, we will switch to WWAN when finally the average RSSI is less than S VIRTUAL CONNECTIVITY MANAGER It is known that when a mobile node moves, its IP address may change, hence the existing connections will be broken. By using the end-to-end approach, which is network infrastructure independent, the mobile users can maintain their on-going connections even if the physical connections have been changed. That is why our scheme called virtual connectivity. The main idea of VC can be illustrated by the following example shown in Figure 4. Subscribe/Notify service Peer Internet ISP2 Peer1 ISP1 Peer WLAN Handoff area GPRS Connection update: Data flow: Movement direction: Figure 4. An example to illustrate how VC works. In Figure 4, suppose Peer1 is communicating with Peer2 via GPRS provided by ISP1. After Peer1 moves into a WLAN provided by ISP2, it will notify Peer2 its new connection information (e.g., IP address and port number). Then they can continuously communicate using WLAN, which has much better bandwidth than GPRS. The S/N service is introduced here to overcome the problems brought by NAT and simultaneous movements. More specifically, VC addresses the following problems which are not addressed (or not fully addressed) by the previous end-to-end approaches. Transparent to application Transparent to application is very important for (legacy) UDP applications since these applications generally use the IP address and port number of the received packet to identify which user this packet

10 belongs to. We use the following example to illustrate the problem: suppose mobile node A is communicating with a server B, the IP address of A is IP A and the IP address of B is IP B. The server program buffers the address (and port) of A and uses it as the identity of A in the program. When A moves to a new place and gets a new IP address, see IP A, the following two cases happen: 1) When A sends packet to B using the new IP address IP A, B cannot know that it is a packet from A, thus the packet cannot be forwarded to the right place; 2) When B sends a packet to A, it still use the old IP address IP A. Hence the packet will be delivered to the wrong receiver. The technique we use to solution this problem is LCT, which will be described later. Transparent to NAT We note that NAT may break the end-to-end mobility under the following scenario: suppose nodes A and B are communicating, node A is behind the NAT box and B is in the public domain. If B moves to a new place and gets a new IP address, it will not be able to notify A its new connection information due to the separation of the NAT box, hence the connection breaks. Simultaneous movements We also note that the connectivity may be broken under simultaneous movements. If mobile nodes A and B move to new network attach points simultaneously, they cannot know the exact address of their peer. Therefore, the update information cannot be sent to the right. To solve the problem faced by NAT and simultaneous movements, a third party service, S/N service is proposed in this work. Besides LCT and S/N service, we also design a VC protocol to exchange connection related information between the end hosts and between the end host and the S/N service. This VC protocol includes several key components, such as peer negotiation, connection maintenance, and S/N protocol. Due to the space limitation, we will not describe them in detail. In the following, we describe how LCT and S/N service work. 4.1 Local Connection Translation LCT is the technique we use to address the application transparency problem. The idea is to maintain a mapping relationship between the original connection information (e.g., IP address and port number) and the current connection information for each active connection as illustrated in Figure 5(a). The original connection information does not change during the connection life time; whereas the current connection information changes each time the node or the peer gets new IP address. The upper layer applications only see the original connection information. Therefore, the mobility is transparent to the upper layer applications.

11 send org_saddr, org_sport, org_daddr, org_dport, proto cur_saddr, cur_sport, cur_daddr, cur_dport, proto recv (a) Local Connection Translation in VC. B:IP B 2 B:IP B 3 CU data CUR/CU 5 3 S/N service CU A:IP A 2 A:IP A NTY B->IP B NAT NAT (b) Subscribe/Notify service in VC. Figure 5. Two key components in VC. The following actions are performed by VC when the node sends and receives packets. Packet sending. When application sends a packet, VC looks up the table to substitute the original connection information to the current one, and then deliver the packet to the lower layer. Packet receiving. When VC receives a packet from outside, it also looks up the table to substitute the current connection information of the packet to the original packet information, and then deliver the packet to the upper layer. The mapping item is created when a new connection is established, and the item is updated when the source and/or destination nodes move to new network access points. 4.2 The Subscribe/Notify service The S/N service is introduced to overcome the problems (i.e., under NAT and simultaneous movements) that cannot be solved by the traditional end-to-end approaches. The idea is to introduce a third party S/N service to bridge the two communication parties together so that they can exchange connection information via this S/N service to resume their otherwise broken end-to-end connection. We assume S/N server is public addressed and never moves.

12 As to the previous NAT scenario, both A and B are connected to an S/N server, and A subscribes to B s IP address changes via the S/N server. If B (the node with public address) moves to a new network access point, it will inform S/N its new address, and then S/N notifies A of this new address of B. In this way, A can resume the connection with B. As to the simultaneous movements scenario, both A and B are connected to an S/N server, A and B subscribe to the IP address changes of each other; so when A and B moves simultaneously, they inform the S/N of their new IP addresses, and then S/N server notifies A and B of the new address of other peer. After that, A and B can resume the connection. Here we use the example illustrated in Figure 5(b) to describe how S/N works to support simultaneous movements and under NAT scenario. Note that both A and B maintain a connection with the S/N server. The connection between A and B are maintained using the following steps. 1) Mobile nodes A and B are communicating with each other, A with private address IP A and B with public address IP B ; 2) A and B moves simultaneously and get new IP address IP A and IP B, respectively; 3) A and B send connection update (CU) message to S/N to update the connection with S/N, and A also sends CU to IP B to update the connection with B, this message will be lost since B has moved away (B does not send CU to IP A since A is private addressed); 4) S/N notifies A of B s new IP address via a notify message (NTY B->IP B ); 5) A issues connection update request (CUR) message to B s new IP address and B replies a CU message to A s new IP address; 6) The connection between A and B is resumed. 5 PERFORMANCE EVALUATION To demonstrate the feasibility and effectiveness of our approach, we have designed and implemented a prototyping system based on the architecture as illustrated in Figure 6 in the Windows 2000 operating system. In our implementation, CM manipulates and monitors all the wireless network interfaces via the NDIS device interface, and provides related information (WLAN_WEAK and WLAN_AVAILABLE, available bandwidth, delay, etc.) to the roaming decision maker, which runs in the system as a background service. The decision maker then triggers VC to perform connection maintenance functionalities when the roaming condition can be satisfied. VC is implemented in the system together with the TCP/IP stack as the tcpip6.sys system driver since it locates naturally between the network and the transport layers. The protocol stack we use in this implement is the IPv6 stack from Microsoft Research.

13 Ethernet Desktop B Transport Layer WLAN AP1 AP2 Virtual Connectivity Manager Legacy IP Layer Notebook A Connection Manager GPRS Link Layer Figure 6. Network setup of the experiment. We use the following experiment to demonstrate the ability of our system to maintain the on-going user connection (TCP in this case) and the ability of our system to achieve better performance (higher user perceived TCP throughput in our case) against the traditional scheme. The basic network setup for our experiment is shown in Figure 6. Specifically, there are two Wireless LAN access points connected to the same Ethernet segment. The mobile node A has installed an IEEE b WLAN card and a GPRS card, whereas the desktop PC B has an Ethernet card. The GPRS network (provided by China Mobile) is always on. That is to say, node A can always maintain the connection with node B via the GPRS network even if the WLAN is not available. Since there exist other WLAN APs during the testing, therefore the interference between other APs is not negligible in the experiment. There is an always on TCP session between A and B to transmit data as fast as possible. For comparison, we also implement the existing roaming approach. In the traditional approach (TA for abbreviation) as used in [11], a handoff from WWAN to WLAN will be triggered as long as the WLAN RSSI is larger than a threshold; likewise, a handoff from WLAN to WWAN will be triggered once the WLAN RSSI is smaller than a threshold. According to the IEEE b standard requirement, we fix the threshold in the TA to be -76dB. During the experiment, the user with a notebook first moves around within the WLAN coverage area covered by multiple APs and then steps out of the WLAN. The connection may be switched between AP1 and AP2 in WLAN at the link layer or between WLAN and GPRS at the network layer. The switch at the link layer is transparent to VC and does not trigger VC actions. We compare the performance of the traditional solution (TA + VC) and our solution (CM + VC). In TA+VC, even we move around within the WLAN, several vertical handoffs from WLAN to WWAN or WWAN to WLAN may be triggered due to the severe RSSI fluctuation and slow intra-wlan handoff. Figure 7 shows the TCP throughput comparison of the connection between A and B under the two approaches.

14 Figure 7. Throughput comparison of the TCP connections. During the whole testing procedure, TCP connection is maintained successfully under both cases due to the effect of VC. From Figure 7 we can see that the using CM+VC, people can obtain much higher throughput since CM can notify the handoff decision more accurately. In this experiment, the TA triggers five vertical handoffs (three for WLAN to GPRS and two for GPRS to WLAN), while CM only triggers 1 vertical handoff when finally stepping out of WLAN. The average TCP throughputs are 2.5Mbps and 1.1Mbps under CM+VC and TA+VC, respectively. Note that the GPRS network is of low bit rate, can only achieve approximate 20kbps throughput for TCP. Moreover, due to the low RSSI and unknown interference, sometimes the transient throughput may also be low even in WLAN, but still larger than the one in GPRS. 6. CONCLUSIONS In this paper, a novel mobility management scheme is proposed for seamless roaming between heterogonous networks such as WWAN and WLAN. The system integrates a connection manager that can intelligently detect the wireless networks change and a virtual connectivity manager that can maintain connectivity using the end-to-end principle. The collaboration between CM and VC accomplishes seamless and efficient roaming between WWAN and WLAN. More specifically, in CM, MAC sensing, FFT detection and adaptive threshold configuration algorithms are proposed to significantly reduce the handoff rate and ping-pong effect during handoff procedure. In VC, end-to-end connection maintenance mechanism is used to make it independent of the network infrastructure. Meanwhile, local connection translation and subscribe/notify service introduced in VC provide advantages of application transparency, working under NAT, and successfully handling of simultaneous movements. Our prototyping system

15 demonstrates that seamless roaming between WWAN and WLAN can be achieved and meanwhile much better performance can be obtained comparing with the one using traditional scheme. Acknowledgement Authors thank Dr. Yuqi Yao, Dr. Kun Tan, and Mr. Hongbin Liao for their valuable discussion. Authors also would like to thank our visiting students Jun Yuan and Jian Li for performing the related experiments. REFERENCES [1] G. Corazza, D. Giancristofaro, and F. Santucci, Characterization of Handover Initialization in Cellular Mobile Radio Networks, in Proc. IEEE VTC 94. [2] M. Gudmundson, Analysis of Handover Algorithms, in Proc. IEEE VTC 91. [3] K. Pahlavan, P. Krishnamurthy, A. Hatami, M. Ylianttila, J. Makela, R. Pichna, and J. Vallstrom, Handoff in Hybrid Mobile Data Networks, IEEE Personal Communications, Apr [4] J. McNair, I. Akyildiz, and M. Bender, An Inter-System Handoff Technique for the IMT-2000 System, in Proc. IEEE INFOCOM [5] C. Perkins, Editor, IP Mobility Support for IPv4, RFC 3344, [6] C. Pekins, et al., Mobility Support in IPv6, [7] R. Ramjee, K. Varadhan, L. Salgarelli, S.R. Thuel, Shie-Yuan Wang, and T. La Porta, HAWAII: a domain-based approach for supporting mobility in wide-area wireless networks, IEEE/ACM Trans. Networking, vol. 10, no. 3, June [8] E. Wedlund and H. Schulzrinne, Mobility Support using SIP, in Proc. WoWMoM'99, Seattle, Washington, August [9] A. C. Snoeren and H. Balakrishnan, An End-to-End Approach to Host Mobility, in Proc. Mobicom 00. [10] IEEE standard for wireless LAN-medium access control and physical layer specification, part 11, IEEE Press, New York, NY, [11] EURESCOM Project P1013-FIT-MIP, First steps towards UMTS: Mobile IP services, an European tesbed,