Reactive model correction with application to admission control of VPN services

Abstract

This paper introduces a VPN service centric point of view on the admission control problem. Considering a VPN service that can carry traffic from several classes of service (COS), it is desirable, in order to simplify the service management, to have all of traffic between two end points of the service, regardless of the class of service that it belongs to, follow the same path. In this case, the service traffic requirements are solved together. Thus, this represents a generalization of the connection admission control problem to multi class of service requests. It is known that real network traffic exhibits deterministic temporal characteristics, and as such, it is nonstationary. Therefore an admission control system that operates under nonstationary conditions is highly desirable. Our approach to this problem is to use a model correction technique, which allows us to update dynamically the traffic model. The technique is based on a loss estimator for aggregated traffic, in which the unknown is one of the traffic model parameters, usually either the variance or the Hurst parameter, while the loss itself is obtained by direct measurement This approach is feasible for high speed links, where loss data can be obtained within a short time period, even for relatively low loss targets.