An Efficient Path Precomputation Technique for QoS Routing Hanan H. Elazhary, Swapna S. Gokhale and Reda A. Ammar Dept. of Computer Science and Engineering University of Connecticut, Storrs, CT 06269 Email: {hanan,ssg,reda}@engr.uconn.edu Abstract To enable QoS routing it is necessary to compute the QoS metrics of the paths through the network. Path precompu- tation, which consists of computing the QoS metrics of the paths at regular intervals can be used for this purpose. The precomputed paths are then used to select a suitable one to route a requested QoS connection. The primary drawback of path precomputation is that the precomputed paths may not be an accurate depiction of the present network state when they are used for routing, which may lead to inef- ficient and inaccurate routing decisions. The discrepancy between the precomputed paths and the current network state increases as the time elapsed from the last precom- putation process increases. In order to alleviate this issue, path precomputation must be performed frequently and to facilitate frequent precomputation, the communication and the computation overheads associated with path precompu- tation must be reduced. In this paper we present an efficient path precomputation technique which reduces the overheads associated with path precomputation significantly over the prevalent precompu- tation techniques. We compare the overheads of the pro- posed technique with the prevalent techniques via extensive simulations. Since the proposed technique offers a signif- icant reduction in the path precomputation overheads, for given level of overheads, it can be employed more frequently than the prevalent techniques. Intuitively, this suggests that the quality and the efficiency of the routing decisions facil- itated by the proposed technique will be better compared to the prevalent techniques for the same level of overheads. Our simulation results confirm this intuition. 1 Introduction The present best-effort Internet architecture is not suit- able for soft, real-time multimedia flows which are charac- terized by high-speed continuous traffic and whose packets should arrive in order and in real time. Such multimedia flows require connections with QoS requirements, which consist of constraints on parameters such as packet delay, packet loss, and connection bandwidth. An important aspect of providing QoS to multimedia flows is QoS routing, which is concerned with determining paths through the network that have sufficient resources to satisfy the QoS requirements of a connection, while simul- taneously achieving global efficiency in network resource utilization [10]. QoS routing requires that the state of the links in the network be represented using a set of QoS met- rics such as bandwidth, delay, delay jitter and cost. In order to determine suitable paths, the QoS metrics of the paths through the network need to be composed using the QoS metrics of the links along each path. This process of com- puting the QoS metrics of the paths is referred to as path computation [3]. Paths can be computed on-demand, upon the arrival of a QoS connection request. On-demand path computation is not scalable and may also incur a long pre- computation delay. Path precomputation, which involves computing the QoS metrics of the paths at predetermined intervals and not in response to any requested QoS con- nection is an attractive alternative to alleviate the scalability and the delay issues associated with on-demand path com- putation [10, 4, 11]. When a QoS connection needs to be routed, the precomputed paths are then used to select a suit- able path. In a dynamic network, where connections arrive and depart often, the precomputed paths may not reflect the present state of the network when they are used for routing. Such discrepancy between the present state of the network and the state reflected by the precomputed paths increases as the time elapsed from the last precomputation cycle and the time when the paths are used for routing increases. This discrepancy may result in inefficient routing decisions or in the worse case an unnecessary rejection of some con- nections. To improve the possibility that the precomputed paths are a close representation of the network state at the time they are used to make a routing decision, path pre- computation must be performed frequently. However, the precomputation process incurs communication and compu- tation overheads, and these overheads increase with an in- 1 2006 IEEE International Symposium on Signal Processing and Information Technology 0-7803-9754-1/06/$20.00©2006 IEEE 714