Prediction-based routing for cell-switched networks

ATM is the target switching and multiplexing technology for implementation of B-ISDN. However, existing ATM software systems are deficient in many respects, such as QoS-based routing. Noting that the route of a virtual circuit remains fixed throughout a session, it is clear that route selection will have long-term effects on network congestion, especially for long sessions. Hence, it is important that routing decisions be based not only on current network state information, but also on predicted future network state information. In this paper we investigate a new prediction-based routing methodology. An ATM LAN software simulator, SimATM, is developed and utilized in this research. SimATM is used to investigate the relative performances of various routing algorithms and link weight assignment strategies. The results clearly indicate that prediction-based routing significantly improves performance with respect to average cell delay, throughput, and jitter. Furthermore, this improvement is shown not to come at the expense of background traffic performance.