Routing over Interconnected Heterogeneous Wireless Networks with ...

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2008 proceedings.

Routing over Interconnected Heterogeneous Wireless Networks with Intermittent Connections Hany Samuel and Weihua Zhuang

Bruno Preiss

Department of Electrical and Computer Engineering University of Waterloo Waterloo, Ontario, Canada N2L 3G1 Email:{hsamuel, wzhuang}@bbcr.uwaterloo.ca

Research In Motion Limited Waterloo, Ontario, Canada N2L 5Z5 Email: [email protected]

Abstract—The recent years have seen an enormous advance in wireless communication technology and a wide spread of various types of wireless networks. It requires effective inter-networking among the heterogeneous wireless networks in order to support user roaming over the networks while maintaining the connectivity. One of main challenges to achieve the connectivity over a heterogeneous wireless network is potential intermittent connections caused by user roaming. The issue is how to maintain the connection as the user roams and how to ensure service quality in the presence of a long disconnection period. In this paper, we apply the concept of delay tolerant network (DTN) framework to heterogeneous terrestrial wireless networks, and propose a routing scheme to ensure successful communication over an information transport platform that can encounter excessive long delays and intermittent paths. Simulation results demonstrate that our proposed scheme outperforms the epidemic routing based scheme. Index Terms—Routing, delay tolerant network (DTN), heterogeneous wireless networks, connectivity, intermittent links, user mobility.

I. I NTRODUCTION Over the recent years, wireless technology has been gaining the momentum with a wide spread of the wireless networks. Various types of wireless networks have been deployed and widely used, such as 3rd-generation cellular networks, wireless local area networks (WLANs), sensor networks, and mesh networks. Each type of network is optimized for a specific networking environment, and it is impossible to have only one type of wireless network that suits all the environments. The global information transport platform via internetworking has become an incontrovertible fact, regardless of the differences among these networks either in structure or in protocols used over them. This leads to the existence of heterogeneous wireless networking. How to support user roaming over heterogeneous wireless networks is an area that has been extensively researched (e.g., [1], [2], [3]). The solutions are mainly based on the Mobile IP protocol to deal with user mobility at the network layer. One limitation with these techniques is that, if a user is disconnected for a long period before reconnecting through another network, the current connection is terminated. Another problem is how to maintain the connection for intermittently

connected users even over the same network. How to provide a continuous connection for users that encounter these difficulties is the main issue that we try to address in this paper. We consider a challenged network consisting of interconnected heterogeneous wireless networks. Challenged networks [4] can be defined as the networks that satisfy one or more of the following characteristics: • •

A communication path between a traffic source and its destination may never exist; The time to send a message from the source to the destination is excessive, due to many factors such as user mobility, link failure, vertical handoff between heterogeneous networks, limited bandwidth, and path instability.

Many wireless networks exhibit these properties. However, it is generally assumed, either explicitly or implicitly, that these properties are rarely encountered and, as a result, they can be and have been neglected. The set of protocols and techniques currently used over most wireless networks fail to adapt themselves to any situation characterized with these properties. For example, it is always implicitly assumed for successful communications over a wireless network that an end-to-end path exists between the source and the destination, which violates the first property that this path may never exist. There are some limited works addressing issues in challenged networks, each with its own protocol, which limits the possibility of internetworking between different types of networks without the need of a highly specialized proxy. With the current evolution of wireless network technology, the need for interoperability among various networks becomes obvious. The delay tolerant network (DTN) architecture has been proposed to deal with the issue in space communications. The problem of creating a connection among nodes (e.g. spacecraft) located deeply in space is discussed in [5]. The proposed architecture is based on an Internet-independent middleware, which is a new layer (referred to as Bundle layer). It handles the sending and receiving of the bundles (self contained messages) across the network using the underlying protocols stack and hides the nature of the communications from the upper layers (especially the application layer). The DTN concept is generalized in [4] for all the challenged networks. Due to the expected long delays, DTNs cannot

978-1-4244-2075-9/08/$25.00 ©2008 IEEE


Fig. 2.

Fig. 1.

Heterogeneous wireless networks connected over Internet backbone

accommodate any real-time applications, but mainly support time insensitive applications. One of main challenges in DTNs is routing, where a route can be a virtual path as there may not exist an end-to-end physical path between the source and the destination. Routing in a DTN depends on the expected nature of the contacts in the network, which can be classified as opportunistic contacts [6], scheduled contacts [6]–[8], and predicted contacts [9]–[12]. The idea of using vehicular networks for DTN routing has been introduced [13]–[16]. For example, routing can be scheduled using an autobus network to send the message [7] or using a vehicle that comes on a regular basis to receive or deliver messages [15]. In this paper, we apply the DTN architecture to terrestrial wireless communications, in order to achieve a continuous connection for mobile users over heterogeneous wireless networks that are connected through an Internet backbone. We propose a new scheme for reliable message delivery for users roaming over heterogeneous networks. It is the first time (with the best of our knowledge) that the DTN architecture is being adopted to cope with the user roaming problem in this environment. The remainder of this paper is organized as follows. Section II describes the system model under consideration. Section III discusses the research problem and proposes our solution to maintain user connectivity in the presence of intermittent links, based on the DTN concept. In Section IV, computer simulation results are presented to demonstrate the effectiveness of the proposed solution. Section V concludes this research. II. T HE S YSTEM M ODEL We consider a global information transport platform, which consists of a number of heterogeneous wireless networks (e.g. cellular networks, ad hoc networks, WLANs, etc.) that are interconnected over an Internet backbone [17], as shown in Figure 1. Each network is connected to the Internet through a DTN gateway [4]. The communication between the gateways is reliable over the Internet backbone. Here, we focus on data communications for delay insensitive applications.

Illustration of the Store and forward message delivery mechanism

Each mobile node is able to connect to the platform through a subset of the wireless access networks. A node may be connected for a period through one access network, disappear for an extended period, and then reappear from the same access network or from a different access network. The user connection over the access networks can be intermittent with frequent disconnections. To deal with the potential long delays encountered in the presence of frequent disconnections, the only effective way for successful message delivery is to use an asynchronous message forwarding mechanism, also known as store and forward mechanism. Consider a simple example as shown in Figure 2. At time T1 , node A wants to send a message to node B, but there is no path between nodes A and B. Node C is a mobile node which can communicate with node A at time T1 and with node B at time T2 , where T2 > T1 as node C moves away from node A toward node B. Let node A send the message to node C at time T1 , node C store the message, and then forward it to node B at time T2 . In this way, node C works as an intermediate node between nodes A and B. The unique feature here is that the path A → C → B is a virtual path that does not exist in its entirety at any one instant in time. The key to a successful communication in this scenario is that, node C stores the message until a right time to forward it to the destination or to another intermediate node. III. T HE P ROPOSED ROUTING S CHEME To maintain communication with a user that encounters intermittent connections in the system (e.g., when roaming over multiple heterogeneous networks), we employ the asynchronous message transfer mechanism as in the DTN architecture with the bundle layer. For a large heterogeneous interconnected wireless network where users are allowed to connect through any of the access networks, our main research issue is how to find a route from a traffic source node to the destination node to deliver the message even if the route does not really exist at any time. Our objectives in finding a routing solution are summarized as follows. • Message delivery: Achieving successful message delivery is the main goal, which poses significant technical challenges for the system under consideration. • Efficient resources utilization: There is always a tradeoff between efficient resource utilization and improving the message delivery probability. In general, routing should


achieve best message delivery rate with minimum resources usage. The resources include node buffer space and power, and the system radio bandwidth. Two key issues should be investigated for the routing: • User location - As a user roams, the node location is not fixed. Thus, the first challenge is to locate the destination node in order to communicate with it. • Message storage - With the expected disconnectivity and/or the possibility of no existing path, the messages need to be stored at some node(s) and forwarded to the destination whenever possible. In the following, we first discuss two possible solutions, and then propose our main scheme based on a concept of super nodes. A. Epidemic Routing Based Scheme This is a distributed routing approach. As no assumption is made on information of user location, epidemic routing [6] is applied. As shown in Figure 1, if a node (node A) wishes to send a message to another node (node B), according the epidemic routing, node A should forward the message to all the nodes it encounters, which continue to forward the message to other nodes, and the procedure stops when the message reaches the destination node. The situation can be simplified: Node A first forwards the message to the DTN gateway of its access network, then the gateway forwards the message to all the connected gateways using the Internet backbone, and finally each gateway forwards the message within its own network. The problem with this approach is that, for a large network size, the number of forwarded messages required to deliver one message is huge. The message flooding in the network results in a large number of lost messages (as to be discussed in Section IV). B. Centralized Node Scheme This is a centralized routing approach. A server resides in the Internet (as shown in Figure 1), and is able to communicate with all the gateways. Every user upon connecting to the system must inform the server of its current location. When a node wants to send a message, it first contacts the server to find out where the destination node is located, then the source node can try to establish a direct connection with the destination node. If the path setup fails or the connection drops at any time, all the messages are sent to and stored at the server for retrieval from the destination node upon reconnecting. The main problem with this solution is scalability. The server becomes a bottleneck in the system even for small-size networks. For large-size networks, this solution is not practical at all. C. New Architecture with Super Nodes We propose to combine the preceding two approaches to overcome their limitations. Instead of having a single server, a number of servers at fixed locations, referred to as super nodes, are used. Each super node is responsible for a set of subscribers. Each user (mobile node) has a unique and fixed super node, independent of its location changes. The

Fig. 3.

The proposed network architecture with super nodes

communication between a super node and a DTN gateway is assumed to be reliable over the Internet. In fact, a super node can also act as a DTN gateway; in this case, the super node is responsible for message delivery over the wireless access networks for which it acts as a DTN gateway. Each mobile node should contact its super node to update its location upon connecting to any access network. To send a message, the source node first locates the super node of the destination node based on the user ID hashing, using a technique such as CAN [18]. With the most updated location of the destination node provided by its super node, the source node then tries to establish an end-to-end connection with the destination. If the connection setup fails or the connection drops at any time, all the messages are sent to and stored at the destination super node for forwarding to the destination node upon its reconnection. Depending on the current access network of the destination node, its super node may transfer the user message custody to another super node near the access network for the period of user existence in the access network. This super node is called custodian super node for the user during the period. To better illustrate the approach, consider a simple scenario as shown in Figure 3. Node A wants to send a message to node B. By hashing the ID of node B, it locates the super node SB of node B. Node A sends to SB a query about node B’s location. Super node SB sends to node A a message that contains the last known location and the custodian super node (the current super node on custody of node B messages) of node B. Then, node A tries to establish a direct connection to node B, that may be possible in some cases (e.g., if node B is connected through a WLAN or a cellular network) or may be infeasible (e.g, if node B is connected through a sparse ad hoc network with an intermittent link). Suppose that node B is connected through the wireless ad hoc network and its current custodian super node is SD . Node A first tries to establish a connection with node B directly, but fails; then node A sends the messages to super node SD . It is SD ’s responsibility to deliver the messages to node B over the access network. If


Fig. 4.

The total number of exchanged messages.

node B is disconnected from an access network in custody by super node SD , then SD will send the stored messages back to node B’s permanent super node SB . Afterward, super node SB updates its record, declaring itself the current custodian super node for B. When node B informs SB of its current network upon its reconnection, super node SB will transfer the stored messages and custody to the new custodian super node of node B based on node B’s current access network. Due to the expected intermittent connection between node B and its current access network, there may be bounced messages due to the custody transfer. To prevent message bouncing, the custodian super node does not transfer the custody until a timeout period that depends on the network nature and the load on that custodian. If the custody is not transfered and SB finds that B is connected via an access network not in jurisdiction of its current custodian super node, its permanent super node SB will inform the current custodian to transfer the custody to the new custodian (i.e. the super node in custody for the new access network). It should be noted that the number of super nodes in the system is not dependent on the number of the access networks (unless a super node also functions as the gateway of the access network). The number of the super node is a function of the number of the users in the system. The centralized node scheme can be regarded as a special case of the super node scheme that suits a very small number of users. IV. P ERFORMANCE E VALUATION We compare the performance of the super node scheme with the epidemic routing based scheme. The performance is measured in terms of the number of exchanged messages over the network to capture how each scheme uses the available resources such as radio bandwidth, and the number of undelivered messages to indicate how successful the scheme is in delivering messages. In our experiments, the simulation proceeds in steps. At each step (simulation time), a mobile node can send a new message or handover to a new access

Fig. 5.

The number of undelivered messages

network. The system has 4 super nodes, 5 wireless access networks (i.e., 3 sparse ad hoc networks, 1 WLAN, and 1 cellular network) each connected via a gateway to the Internet backbone, and 50 or 100 mobile nodes distributed randomly within the access networks. Each super node is assigned an (approximately) equal number of users. A user can roam over all the access networks, the residence time over each network being an exponentially distributed random variable with mean of 10 simulation steps. Each user is disconnected for a random period of time (which is an exponential random variable with mean of 10 simulation steps), then reconnected either from the same access network (with probability 0.7) or from any other access network equally likely. All the messages are equal in size, with the same message live time of 4 simulation steps. The buffer space is 20 messages at each mobile node and 2000 messages at each super node. For simplicity in simulation, we use the epidemic routing over ad hoc networks. For each experiment, a communication scenario (i.e., set of messages, user connections, user disconnections, user movements) is set up randomly and run for each scheme. Figure 4 shows the total number of message exchanges for 50 and 100 mobile nodes respectively. It is clear that the super node scheme outperforms the epidemic routing based scheme, requiring a much smaller overhead for message exchanges for successful message transfer. Different from the epidemic routing based scheme, the super node scheme does not need to send the messages over all the networks, but only to the current access network of the destination node. It is noted that the performance improvement increases with the network sizes (represented by the number of mobile nodes here). However, the super node scheme still requires many message exchanges. It is because the majority of the access networks are ad hoc networks, and message delivery over an ad hoc network requires the exchange of many messages when using the epidemic routing within each ad hoc network. Figure 5 shows the number of undelivered messages. It is observed that the super node scheme is much better than the


epidemic routing based scheme. The main problem with the epidemic routing based scheme is that the user messages are forwarded to many other intermediate mobile nodes which may never meet the destination node. Moreover, other intermediate nodes (which will meet the destination node) may have to drop the carried messages due to the buffer space limitation, resulting in lost messages. It is obvious from the figure that the number of lost messages is greatly affected by the number of the mobile nodes in the system. On the other hand, for the super node scheme, the messages are kept at the super node buffer until the destination nodes are reconnected. Hence, message loss is mainly due to the message live time (i.e., the message expired before its destination node is reconnected to the network). The longer the message live time, the smaller the number of the lost messages, but the larger the buffer space needed at the super nodes. The change of the network size (as the number of mobile nodes is increased from 50 to 100) only slightly increases the number of lost messages, because it mainly depends on the super node buffer size but not the buffer size at each mobile node. V. C ONCLUSIONS AND F UTURE WORK In this paper, we investigate how to ensure connectivity to roaming users with intermittent connections over heterogeneous terrestrial wireless access networks. Based on the storeand-forward strategy used in DTN, we propose a routing solution using the super node architecture to achieve reliable communications. Computer simulation results demonstrate that the super node scheme outperforms the epidemic routing based scheme especially for large networks, in terms of message exchange overhead and the number of lost messages. This paper introduces the main idea of the super node scheme. Further work on the super node scheme includes various aspects: one aspect is the routing between the super node and the user over the access network. This routing is subject to the characteristics of the network. Even though in this paper we used epidemic routing over ad hoc networks, using super nodes allows us to gather information of mobile nodes over different access networks, the information can facilitates more efficient routing strategies. Other important aspects are load balancing among super nodes, system scalability, reliability, and system security: in terms of how to enforce access control paradigm for the system resources and how to ensure the message privacy that neither the super node or any intermediate node can expose the message contents. ACKNOWLEDGMENT The authors would like to thank Allan Lewis of the Research In Motion (RIM) for many helpful discussions on this research. This research was supported by research grants from the Natural Science and Engineering Research Council (NSERC) of Canada and from RIM. R EFERENCES [1] M. Shi, H. Rutagemwa, X. Shen, J. Mark, and A. Saleh, “A serviceagent-based roaming architecture for WLAN/Cellular integrated networks,” IEEE Trans. Vehicular Technology, vol. 56, no. 5, pp. 3168– 3181, Sept. 2007.

[2] H. Rutagemwa, S. Pack, X. Shen, and J. Mark, “Robust cross-layer design of wireless profiled TCP mobile receiver for vertical handover,” IEEE Trans. Vehicular Technology, vol. 56, no. 6, pp. 3899–3911, Nov. 2007. [3] C. Guo, Z. Guo, Q. Zhang, and W. Zhu, “A seamless and proactive end-to-end mobility solution for roaming across heterogeneous wireless networks,” IEEE Journal on Selected Areas in Communications, vol. 22, pp. 834–848, June 2004. [4] K. Fall, “A delay-tolerant network architecture for challenged internets,” in SIGCOMM ’03: Proceedings of the 2003 conference on Applications, technologies, architectures, and protocols for computer communications. New York, NY, USA: ACM Press, 2003, pp. 27–34. [5] S. Burleigh, A. Hooke, L. Torgerson, K. Fall, V. Cerf, B. Durst, K. Scott, and H. Weiss, “Delay-tolerant networking: an approach to interplanetary internet,” IEEE Communications Magazine, vol. 41, no. 6, pp. 128–136, June 2003. [6] A. Vahdat and D. Becker, “Epidemic routing for partially connected ad hoc networks,” April 2000. [Online]. Available: citeseer.ist.psu.edu/ vahdat00epidemic.html [7] S. Jain, K. Fall, and R. Patra, “Routing in a delay tolerant network,” in SIGCOMM ’04: Proceedings of the 2004 conference on Applications, technologies, architectures, and protocols for computer communications. New York, NY, USA: ACM Press, 2004, pp. 145–158. ¨ ur B. Akan, C. Chen, J. Fang, and W. Su, “Inter[8] I. F. Akyildiz, Ozg¨ planetary internet: state-of-the-art and research challenges,” Computer. Networks, vol. 43, no. 2, pp. 75–112, 2003. [9] A. Lindgren, A. Doria, and O. Schel´en, “Probabilistic routing in intermittently connected networks,” SIGMOBILE Mobile Computing and Communications Review, vol. 7, no. 3, pp. 19–20, 2003. [10] X. Chen and A. L. Murphy, “Enabling disconnected transitive communication in mobile ad hoc networks,” in Workshop on Principles of Mobile Computing, 2001, pp. 21–23. [11] J. Ghosh, H. Q. Ngo, and C. Qiao, “Mobility profile based routing within intermittently connected mobile ad hoc networks (ICMAN),” in IWCMC ’06: Proceeding of the 2006 international conference on Communications and mobile computing. New York, NY, USA: ACM Press, 2006, pp. 551–556. [12] J. Leguay, T. Friedman, and V. Conan, “DTN routing in a mobility pattern space,” in WDTN ’05: Proceeding of the 2005 ACM SIGCOMM workshop on Delay-tolerant networking. New York, NY, USA: ACM Press, 2005, pp. 276–283. [13] R. Shah, S. Roy, S. Jain, and W. Brunette, “Data mules: Modeling a three-tier architecture for sparse sensor networks,” in Sensor Network Protocols and Applications, 2003. Proceedings of the First IEEE International Workshop on Sensor Network Protocols and Applications. Washington, DC, USA: IEEE Computer Society, 2003, pp. 30–41. [14] W. Zhao and M. H. Ammar, “Message ferrying: Proactive routing in highly-partitioned wireless ad hoc networks,” in FTDCS ’03: Proceedings of the The Ninth IEEE Workshop on Future Trends of Distributed Computing Systems (FTDCS’03). Washington, DC, USA: IEEE Computer Society, 2003, pp. 308–314. [15] W. Zhao, M. Ammar, and E. Zegura, “A message ferrying approach for data delivery in sparse mobile ad hoc networks,” in MobiHoc ’04: Proceedings of the 5th ACM international symposium on Mobile ad hoc networking and computing. New York, NY, USA: ACM Press, 2004, pp. 187–198. [16] J. LeBrun, C.-N. Chuah, D. Ghosal, and M. Zhang, “Knowledge-based opportunistic forwarding in vehicular wireless ad hoc networks,” in VTC 2005-Spring Vehicular Technology Conference, 2005 IEEE 61st, vol. 4, 2005, pp. 2289– 2293. [17] H. Jiang, W. Zhuang, and X. Shen, “Cross-layer design for resource allocation in 3g wireless networks and beyond,” IEEE Communications Magazine, vol. 43, no. 12, pp. 20–126, Dec. 2005. [18] S. Ratnasamy, P. Francis, M. Handley, R. Karp, and S. Schenker, “A scalable content-addressable network,” in SIGCOMM ’01: Proceedings of the 2001 conference on Applications, technologies, architectures, and protocols for computer communications. New York, NY, USA: ACM Press, 2001, pp. 161–172.