SIMULATION STUDY OF CONNECTION PREEMPTION APPROACHES IN DECENTRALIZED NETWORKS

Abstract

Performance study of two different approaches to preempt the low priority connections for providing the reliable services to high priority connections is the main focus of this dissertation. Connection preemption provides available and reliable services to high_priority connections when a network is heavily loaded and connection request arrival patterns are unknown, or when the network experiences link or node failures. Coupled with the capability to reroute connections that are preempted due to failure or preemption, . connection preemption allows a high quality of service to be provided to network connections, and bandwidth to be used more efficiently. It presents a comprehensive simulation study of preemption in a general connection-oriented network setting. Our simulation study also provides useful insights into connection preemption and network dimensioning problems in order to achieve a desired level of network availability. Based on the observations made in this study, we designed two connection preemption selection algorithms that that operate in a decentralized/distributed network where individual link managers run the algorithm for connection preemption selection on their outgoing links. The first algorithm optimizes the criteria of (a) the bandwidth to be preempted, (b) the priority of connections to preempted, and (c) the number of connections to be preempted, in that order, and has exponential complexity. The second algorithm optimizes the criteria of (a) the number of connections to be preempted (b) the bandwidth to be preempted, and (c) the priority of connections to be preempted, in that order, and has polynomial complexity. From a comparison study of these two algorithms we conclude that polynomial algorithm is almost as good as the exponential algorithm in terms of overall network performance.