End-to-End Proportional Differentiation Over OBS Networks

In this paper, we propose a novel scheme to provide end-to-end proportional differentiated services to an arbitrary number of traffic classes at the packet level. The service classes are defined in terms of the packet loss probability measured between the ingress node and the egress node of an OBS network, where each ingress node aggregates a large number of IP flows. Our solution requires only that the OBS network is able to provide relative differentiation to two types of bursts, a task which can be accomplished in very different ways. In order to demonstrate the feasibility and the performance of the proposal, we develop a mathematical model for computing the loss probabilities in multiservice OBS networks. Specifically, we use a fixed-point model and show how to use its results to derive the desired packet loss probabilities. The second contribution of this work is to study the effects of coupling the packet service classes with reactive sources, namely, sources responsive to congestion. In particular, the well-known dynamics of TCP is embodied into the analytical model. The numerical results produced by this analytical framework show that good proportional differentiation, both in packet loss and in throughput, can be effectively achieved without sacrificing bandwidth usage.