A SIP-Based Seamless-Handoff (S-SIP) Scheme ... - Semantic Scholar

3 downloads 17 Views 345KB Size Report
and capable of providing “Always Best Connected (ABC)” .... domain. Once the MN's network address has changed, the UA will register its current address with ...

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

A SIP-based Seamless-handoff (S-SIP) Scheme for Heterogeneous Mobile Networks Jie Zhang1, Henry C. B. Chan2 and Victor C. M. Leung1 1

Department of Electrical & Computer Engineering, The University of British Columbia, Vancouver, Canada 2 Department of Computing, The Hong Kong Polytechnic University, Hong Kong, China Emails: [email protected], [email protected] and [email protected]

Abstract —One of the prominent goals of next-generation mobile communication systems is to support seamless roaming of users across heterogeneous wireless access networks. Although SIP has been regarded as an attractive candidate to provide mobility management in B3G/4G networks because of its flexibility and ease of deployment, it still suffers from unacceptably long handoff delays. In this paper, a SIP-based seamless-handoff (SSIP) scheme is proposed to overcome this problem in interdomain roaming. Unlike the other SIP-based handoff methods, SSIP employs a “make-before-break” procedure for seamless handoffs. Moreover, we also propose the appropriate handoff triggering point for this scheme based on location tracking techniques. Simulations results show that the proposed scheme greatly outperform the general SIP-based handoff scheme.

I. INTRODUCTION Next-generation (NG) wireless networks are expected to be composed of various heterogeneous wireless access networks and capable of providing “Always Best Connected (ABC)” services to mobile terminals equipped with multiple radio interfaces [1][2]. To meet this goal, one of the most important issues is how to support seamless roaming of users across heterogeneous wireless networks. So far, numerous mobility management solutions operating in different layers have been proposed [3][4], such as the Mobile Internet Protocol (MIP) in the network layer [5][6], the Stream Control Transmission Protocol (SCTP) in the transport layer [7] and the Session Initiation Protocol (SIP) [8] in the application layer [9][10]. Among these solutions, the SIP-based solution can achieve true end-to-end mobility management without the need to modify the network architecture or end-user terminals [4]. Moreover, SIP has excellent extensibility and scalability due to its operation at the highest layer and use of text-based control messages. Besides, SIP is also the signaling protocol used for session control in the IP Multimedia Subsystem (IMS) [11] for mobile networks. Therefore, SIP is considered as an attractive candidate to support mobility in NG networks. Although SIP is free of the drawbacks of lower-layer solutions, it incurs considerable handoff delays due to the exchange of application layer messages. It has been proved that such delay is unacceptable for real-time multimedia services [10][12]. Many enhanced SIP-based schemes have recently been This work was supported by a grant from the Bell University Laboratories program, the Canadian Natural Sciences and Engineering Research Council under grant CRDPJ 328202-05, and by the Department of Computing, The Hong Kong Polytechnic University under account number Z09Z.

proposed [13-17] to reduce the handoff delays. For example, hierarchical registration mechanisms have been proposed in [13] to localize signaling messages for intra-domain roaming. A fast handoff method has also proposed in [14], where inflight packets are forwarded to the mobile node (MN) by means of Real-time Transport Protocol (RTP) in the old subnet. Similarly, hierarchical mobile SIP (HMSIP) in [15] handles the mobility with a two-layer registration mechanism. The above schemes focus on intra-domain mobility management, where all subnets have similar characteristics so that they can cooperate easily. Generally, it is more difficult to reduce interdomain handoff delay for the following reasons. First, the underlying domains may be heterogeneous, thus an end-to-end session needs to be re-established during handoff. Second, inter-domain handoff involves more procedures such as reauthentication so the handoff delay is much longer. To address this issue, [16] proposes to establish security associations (SA) between the MN and neighboring domains in advance, and execute the authentication procedure locally so that handoff delay can be shortened. However, this scheme requires modifications to the existing authentication systems of all domains and also incurs heavy signaling costs to distribute the SA messages. A seamless handoff scheme is proposed in [17], where a temporary session between the MN and the old base station (BS) is set up to forward in-flight data packets during the handoff process. However, this scheme requires all BSs in the networks to be equipped with the Back-to-Back User Agents (B2BUA), which may not be preferred by some operators. In this paper, we introduce a SIP-based end-to-end seamless handoff scheme (S-SIP) to support seamless interdomain roaming. Basically, this scheme employs a “makebefore-break” handoff procedure to provide seamless handoff management. S-SIP does not require any modifications to network entities and is easy to implement. Moreover, we also propose the appropriate handoff triggering point for this scheme taking advantage of current tracking technologies. Simulations are carried out to demonstrate that the proposed architecture performs efficiently. The rest of the paper is organized as follows. Section II describes the network architecture and Section III introduces the S-SIP scheme. Section IV presents and discusses the simulation results. Section V concludes the paper. II.


In this section, we present the architecture for SIP-based

1525-3511/07/$25.00 ©2007 IEEE


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.







Set Up Session

(1) Invite (2) OK (3) ACK Media

Handoff Triggering

Attach to New Domain

(4) AAA Request (5) AAA Reply (6) DHCP Discover (7) DHCP Offer (8) DHCP Request (9) DHCP ACK

Set Up New Session

(10) Invite, Join, Referred By UA_I (11) OK (12) ACK



In this section, we propose a SIP-based seamless-handoff (S-SIP) scheme to support seamless handoff for inter-domain roaming. Unlike the general SIP [9][10] that provides “breakbefore-make” handoff, the S-SIP offers a “make-before-break” scheme to achieve seamless handoff. By taking advantage of tracking technologies, a MN can predict the potential handoff

End Old Session

mobility management in NG wireless networks. As shown in Fig. 1, a NG network typically consists of many access networks using heterogeneous wireless technologies, such as Universal Mobile Telecommunications System (UMTS) and wireless local area network (WLAN). All these access networks are interconnected over the Internet using IP. The MN is equipped with multiple radio interfaces that enable it to communicate with various networks simultaneously. The MN is also equipped with a SIP user agent (UA) to manage SIP messages on behalf of the user. UAs are identified by SIP uniform resource identifiers (URI) with an email-like address sip: [email protected] Various control messages are defined [8] to manage sessions between UAs. Generally, the exchange of messages between UAs follows a client-server interaction model. To set up a session, the requesting UA client sends an INVITE message to the target UA server. To terminate a session, a BYE message is sent by either UA. Although the MN may roam through multiple domains, it only has one home domain. Once the MN’s network address has changed, the UA will register its current address with the SIP registrar server in the home domain. Therefore other nodes can always track the MN through its home registrar (HR). Generally, such roaming can happen between or during active sessions, and is referred as pre-call mobility and mid-call mobility, respectively. Precall mobility can be easily solved by sending a REGISTER message to the HR. Mid-call mobility is usually enabled by MN sending an INVITE message to the remote corresponding node (CN) notifying the new domain/IP address and session parameters as well as updating the binding in the HR concurrently. Thus mid-call mobility will lead to a certain period of disconnection, where data packets sent during this period are lost. This delay is made up by several latency factors such as authentication latency, IP address acquisition latency and media-redirection latency. It has been shown in [10] that the overall delay incurred by such a handoff scheme is too long to be acceptable for real-time multimedia services. Therefore, a more efficient handoff management scheme is required.



Figure 1. Network Architecture

(13) Bye (14) OK (15) Register (16) OK

Figure 2. Handoff Procedure

and send the JOIN signaling message and establish simultaneous connections via multiple networks in advance. Such a “make-before-break” scheme can dramatically lessen the negative impact of mid-call mobility on user sessions at the cost of consuming a little additional network resources. This section consists of three parts. In part A, we present details of the S-SIP scheme. The handoff latency is analyzed in part B. Part C discusses how to decide the appropriate handoff triggering time based on location tracking technologies. A. Handoff Procedure in S-SIP Scheme In this part, we use an example to illustrate how a MN roams across heterogeneous domains. Fig. 1 shows the network environment. Note that for simplicity, we ignore some intermediate proxy servers in the figure. Initially, a MN with SIP URI: [email protected] accesses a visited domain: domainA.com via an interface UA_A. The MN registers the contact address: [email protected] to the HR server located in its home domain. Then a CN in a remote domain with SIP URI: [email protected] initiates communications with the MN. With the SIP URI of the MN, the CN is aware of the MN’s home domain. It sends an INVITE message to the MN’s SIP server. The SIP server acquires the MN’s contact address from the HR and forwards the INVITE message to the MN. Upon receiving the INVITE message, the MN agrees to establish a session by replying with an OK message. The CN then acknowledges the OK message with an ACK message. After negotiation, a media session is set up to exchange data packets. During the session, the MN detects the beacon of another visited domain B: domainB.com via another interface UA_B. The MN starts to evaluate the services offered by domain B and considers whether to switch to the new domain. The final handoff decision can be made based on many factors, such as the QoS of the new domain, requirements of ongoing applications and signal strengths of the current and the new domains. If the MN decides to initiate a handoff, it will first authenticate with domain B and then acquire IP and/or domain addresses through a specific protocol. In this example, we assume that the IP address in domain B can be assigned by a dynamic host configuration protocol (DHCP) server and the MN’s address in domain B is [email protected] After


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

Among these messages, there are five inter-domain messages, e.g., messages transmitted from the MN to the CN/HR, and the other six messages are locally transmitted within domain B. The delays of these two types of messages are as follows.

Figure 3. Headers of SIP Messages

obtaining the new address, the MN sends an INVITE message with a special JOIN header [18] to the CN via the interface UA_B. This JOIN header contains all the relevant information about the ongoing call; e.g., it sets the parameters like the call ID and tags of the previous connection. In this way, the CN knows that the new connection wants to join the ongoing connection between the MN and CN via interface UA_A. After negotiation on the parameters, another session is established between the MN and CN via interface UA_B. Normally, a twoparty conversation will create a conference session [18]. However, here the CN knows the new participant and the peer node of the conversation are logically the same body because their SIP URIs are the same. Therefore, the CN synchronously sends packets to both interfaces. After the transaction, the MN and CN can communicate through the two connections independently and synchronously and the MN will discard any duplicate RTP packets. When the new connection is set up, MN will send a BYE message to the remote CN to terminate the connection via interface UA_A. It also updates the contact address in the HR with the new address [email protected] Fig. 2 shows the generic signaling flow for the handoff procedure and the corresponding SIP messages (i.e., the headers) are given in Fig. 3. B. Estimating Handoff Period In this part, we derive the handoff period of the proposed scheme. Note that, the data packets sent during this period still reach the MN due to the seamless handoff nature. As shown in Fig. 2, the handoff procedure starts with message 4 and ends with message 16. As the registration to the HR can be executed in parallel with messages 10-14, the handoff time only counts from the message 4 to the message 14. Specially, let Dhandoff be the handoff delay and DA↔B denote the delay of messages transmitted between nodes A and B. We have: (1) Dhandoff = 5DMN↔CN + 2DMN↔AAA + 4DMN↔DHCP

Generally, there are four types of delay: queuing delay, processing delay, transmission delay and propagation delay. The propagation delay over a wireless access network is very small and can be neglected. Basically, the delay incurred by inter-domain messages involves: (a) processing and queuing delay in sending terminals, (b) transmission delay over wireless channels, (c) processing and queuing delay in the BS/AP, (d) delay in the Internet, and (e) queuing delay in receiving terminals. Usually, the processing delay depends on the workload and the serving rate of the node/server. The queuing delay is related to the load of the terminal/server and can be determined based on a queuing model. The transmission delay for a wireless channel is determined by the size of the transmitted message as well as the bandwidth of the wireless channel. The delay on Internet is the time to transmit a message across the Internet. For intra-domain messages, the delay on the Internet should be replaced by the transmission delay between the BS and the local server, whose value is also negligible. Therefore, we can get: D MN → CN = PMN + Q MN + D MN → AAA / DHCP = PMN + Q MN +

L SIP + PBS + Q BS + ∆I + QCN B wireless


L AAA / DHCP + PBS + Q BS + PAAA / DHCP + Q AAA / DHCP B wireless


where Bwireless is the bandwidth of the wireless links and Pi and Qi denotes the processing delay and queuing delay in node i. LSIP, LAAA and LDHCP indicate the length of SIP messages, authentication messages and DHCP application messages, respectively. Similarly, we can also get the value of DCN↔MN and DAAA/DHCP↔MN. Substituting (2) and (3) into (1), we have: Dhandoff = 11PBS + 6PMN + 2PCN + PAAA + 2PDHCP + 5∆I +

5LSIP + 2LAAA + 4LDHCP Bwireless


+11QMN +11QBS + 5QCN + 2QAAA + 4QDHCP

C. Handoff Initiation In the above subsections, we discuss handoff procedures for inter-domain roaming. One of the goals of future NG networks is for the MN to deliver “ABC” services. An active MN detecting multiple available access networks will attach to the one offering the best service. However, whenever the MN moves out of the current domain, it should switch to a less desirable access networks to maintain communications connectivity. Ideally, a handoff should be completed right after the MN moves out of the current domain. This means that a handoff should be initiated when the remaining dwelling period of the MN in the current domain is Dhandoff. In the following, we will discuss how to determine an appropriate time for initiating a handoff. With the decreasing cost of microelectronics and advances in tracking technologies, mobile terminals are increasingly equipped with location-tracking capability (e.g., GPS receiver) with an accuracy of ±1 meter [19]. This capacity is also


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

extremely valuable for mobility management [19][20]. Here we will use this technology to predict the remaining dwelling period of an MN in the domain. To estimate the remaining dwelling period of an MN in the domain, the MN should be aware of the coverage area of the domain and the mobility information. Here, the coverage area of the domain is defined as the area where the radio signal strength (RSS) of the network is strong enough to be detected by the corresponding network interface. Due to the existence of obstacles (e.g., buildings, trees and terrains) which distort the radio wave propagation, the coverage area of the domain may not be circular. We assume that MNs report their positions to the network whenever the RSS falls below a pre-defined threshold. After collecting a number of positioning reports, the network can estimate the coverage area of the domain. Note that the radio wave propagation also varies with the weather and climate conditions. Therefore, computations should be performed periodically to estimate the coverage area. Furthermore, the MN should also predict its future movement. Normally, the future movement is highly related to the past movement. Therefore, the MN should keep track of its location at a regular time interval ∆T (typically 1 s). Similarly, the movement of the MN is also affected by the terrain of the coverage area. For simplicity, we differentiate the coverage area into two categories according to its terrain: open areas where users’ movements are relatively unconstrained, and road areas where users’ movements are constrained by certain directions. In an open area with little obstacles, it should be reasonable to assume that the MN’s movement in the future is solely related to its past movements. Here, denote the past h samples of the recorded positions as (Lxk, Lyk), where k = 0, 2, …, h-1, Lxk and Lyk is the position of MN in both x and y directions, respectively at the previous k-th time unit. Here, the value of h should be set appropriately. If h is too small, the estimation may not be general to reflect the movement of the MN. If h is too large, the estimated value may be insensitive to the change in direction. As the handoff only lasts for several seconds, we can consider the movement trajectory of MN during this period as a straight line. Denote the movement direction and velocity of the MN as θ and v, respectively, we can get the position of terminals in the future t-th time unit as: Lx=Lx0+ vtcosθ and Ly=Ly0+ vtsinθ. The value of v can be estimated as: v=

k = h−2

k =0



k x

− Lkx +1

) + (L 2

k y

− Lky+1



k =h −2




k =0

Here wk denotes the weight on the past k-th time unit which satisfies wk-1>wk>0. The moving direction θ can be estimated by the weighted linear regression as:


θ = arg tan (Ψ′WΨ )−1 Ψ′WY



 L0y − L1y   L0x − L1x      where Ψ =  ...  , Y =  ...  and W=diag{w0, …, wh-2}.  Lh−2 − Lh−1   Lh−2 − Lh−1  y  x   y  x The prediction of the MN’s movement in a road area is similar to that in an open area. The only difference is that θ should be

determined by the road terrains rather than by the past movements. As handoffs take place near the boundary of an access network, the MN should initiate the location tracking function only when the RSS of the access network is lower than a certain threshold. Furthermore, it should also execute the handoff whenever it predicts that it will leave the access network soon. Specifically, as the positions are sensed at regular intervals, the MN should initiate the handoff if it estimates that it will leave the access network within ∆T+Dhandoff time units. Obviously, the prediction may not always be correct. If the MN leaves the access network earlier than the estimated time, the handoff will lead to certain packets losses. If it leaves later than the estimated time, its overall service will be unnecessarily degraded. The balance depends on the traffic type and user’s preference. Consequently, we suggest adding an adjustable factor in the expression of the remaining time threshold (i.e., it becomes µ(∆T+Dhandoff), where the value of µ is chosen by the user). IV.


In this section, we implement the proposed handoff scheme in a simulation model and compare its performance with that of the general SIP-based handoff scheme [9][10]. The network topology of the simulation model is shown in Fig. 1. There are 100 MNs roaming in the networking area, whose movement follows the Gauss-Markov model [20]. Each MN is equipped with a GPS with an accuracy of 1 m. Table 1 shows the values of relevant parameters. Note that the Internet delay (∆I) is determined by many factors (e.g. number of routers) and thus is rather difficult to estimate. Here we set ∆I between 50~200ms. Fig. 4 shows the delay performance for a vertical handoff. Here the Y axis shows the sequence numbers of packets receiving at the corresponding time on X axis. In this figure, the MN leaves domain A at time 4 s. The bandwidth of domains A and B are 1.5 Mbps and 1 Mbps, respectively. From the figure, we can see that the general SIP-based hard-handoff handoff scheme causes a handoff delay of around 1.2 s. In this scheme, the old connection is broken before the new connection is set up and the MN attaches to domain B when it loses the signals from domain A. Comparatively, it is evident that the proposed seamless-handoff scheme supports seamless roaming between domains. At time 2.65 s, the MN predicts a future movement out of the current domain. So it attaches to domain B, obtains a new domain/IP address and joins the session with the new address at about time 3.85 s. Once the TABLE I SIMULATION PARAMETERS Parameters


Radius of WLAN coverage Mean Velocity of MN Velocity Variance of MN ∆T h

100 meter 1 m/s 1 m/s 1 second 5 1 20 ms 200 ms 50ms~200 ms 140 bytes 100 bytes 1 Mbps~2 Mbps




This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.


1200 S-SIP scheme General SIP scheme Connection via interface A Connection via interface B

1400 1200 Average Handoff Delay (ms)

Packet Sequence Number


800 Duplicated Packets 600 Handoff Delay 400

Activate handoff 0




Leave Domain A

4 5 Received Time (s)



Fig. 5 displays the handoff delay in different network configurations. Generally, the handoff is seamless when the new connection is set up before the old connection ends. However, due to the errors on position tracking and movement prediction, the MN may move out of domain A before receiving packets through the new connection. In this case, some data packets may be delayed and the handoff is not seamless. From the figure, we can see that the handoff delay of the proposed scheme is very close to zero, which means most handoffs are seamless. Note that the user can further control the seamlessness of handoff by adjusting the value of the factor µ. Also, the handoff delay of the general SIP scheme all exceeds 1 s, which is unacceptable to real-time multi-media traffic. CONCLUSION

The general SIP-based handoff scheme is popular for its easy deployment but suffers from unacceptably long handoff delays. In this paper, we have proposed a novel SIP-based scheme to support seamless roaming of users between heterogeneous access networks. By sending the JOIN signaling message, the mobile node can simultaneously establish two ongoing connections with the CN and realize the “seamlesshandoff”. Moreover, with the help of tracking technologies, the MN can predict potential handoffs and establish the new connection before the current connection losses its signal. Simulation results show that the proposed scheme can support seamless inter-domain roaming. REFERENCES [2]





MN receives the packets through the new interface, the MN terminates the old connection and leaves domain A at time 4 s. Virtually no packets are lost during this handoff procedure.


S-SIP scheme General SIP scheme Internet delay=50 ms Internet delay=100 ms Internet delay=150 ms



E. Gustafsson and A. Jonsson, “Always best connected”, IEEE Wireless Communications Magazine, vol. 10, no. 1, pp. 49-55, Feb. 2003. V. Gazis, N. Alonistioti and L. Merakos, “Toward A Generic ‘Always Best Connected’ Capability in Integrated WLAN/UMTS Cellular Mobile Networks (And Beyond)”, IEEE Wireless Communications Magazine, vol. 12, no. 3, pp. 20-29, Jun. 2005. I. F. Akyildiz, X. Jiang and S. Mohanty, “A Survey of Mobility Management in Next-Generation All-IP-based Wireless Systems”, IEEE Wireless Communications Magazine, vol. 11, no. 4, pp. 16-28, Aug. 2004.



1.3 1.4 1.5 1.6 1.7 Average Channel Bandwidth (Mbps)




Figure 5. Average Handoff Delay

Figure 4. Handoff Events




New connection established




[5] [6] [7]

[8] [9]


[11] [12]






[18] [19]


N. BanerJee, W. Wu and S. K. Das, “Mobility Support in Wireless Internet”, IEEE Wireless Communications Magazine, vol. 10, no. 5, pp. 54-61, Oct. 2003. C. E. Perkins, “IP mobility support for IPv4”, RFC 3220, Jan. 2002 C. E. Perkins and D. B. Johnson, “Route Optimization in Mobile IP”, Internet draft, IETF, draft-ietf-mobileip-optim-11.txt, Sep. 2001. L. Ma, F. Yu and V. C. M. Leung, “A New Method to Support UMTS/WLAN Vertical Handover Using SCTP”, IEEE Wireless Communications, vol. 11, no. 4, pp. 44-51, Aug. 2004. J. Rosenberg, et al, “SIP: Session Initiation Protocol”, IETF RFC 3261, Jun. 2002. H. Schulzerinne and E. Wedlund, “Application-Layer Mobility Using SIP”, Mobile Computing and Communications Review, vol. 4, no. 3, pp. 47-57, Jul. 2000. W. Wu, N. Banerjee, K. Basu and S. K. Das, “SIP-based Vertical Handoff Between WWANs and WLANs”, IEEE Wireless Communications Magazine, vol. 12, no. 3, pp. 66-72, Jun. 2005. 3GPP, “IP Multimedia Subsystem (IMS)”, TS 23.228 (v 6.10.0), Jun. 2005. N. Banerjee, K. Basu and S. K. Das, “Handoff Delay Analysis in SIPbased Mobility Management in Wireless Networks”, Proc. of the International Parallel and Distrubted Processing Symopsium (IPDPS), pp. 224.2, Apr. 2003. H. Schulzerinne and E. Wedlund, “Application-Layer Mobility Using SIP”, Mobile Computing and Communications Review, vol. 4, no. 3, pp. 47-57, Jul. 2000. A. Dutta, S. Madhanie, W. Chen, O. Altintas and H. Schulzrinne, “Optimized Fast-Handoff Schemes for Application Layer Mobility Management”, ACM Mobicom Poster, Sep. 2002. D. Vali, S. Paskalis, A. Kaloxylos and L. Merakos, “An Efficient MicroMobility Solution for SIP Networks”, Proc. of Globecom, pp. 30883092, Dec. 2003. T. T. Kwon, M. Gerla, S. Das and S. Das, “Mobility Management for VoIP Service: Mobile IP vs. SIP”, IEEE Wireless Communications, vol. 9, no. 5, pp. 66-75, Oct. 2002. N. Banerjee, A. Acharya and S. K. Das, “Seamless SIP-Based Mobility for Multimedia Applications”, IEEE Network Magazine, vol. 20, no. 2, pp. 6-13, Mar.-Apr. 2006. R. Mahy and D. Petrie, “The Session Initiation Protocol (SIP) ‘Join’ Header ”, IETF RFC 3911, Oct. 2004. W. Soh and H. S. Kim, “Dynamic Guard Bandwidth Scheme for Wireless Broadband networks”, Proc. of IEEE INFOCOM’2001, pp. 572-581, Apr. 2001. B. Liang and Z. J. Haas, “Predictive Distance-based Mobility management for multidimensional PCS Networks”, IEEE/ACM Trans. Networking, vol. 11, no. 5, pp. 718-732, Oct. 2003.


Suggest Documents