Network state models for analysis and aggregation in large-scale quality of service -aware multiclass networks.

摘要

The advent of the Internet followed by the accelerated macro-evolution of the World Wide Web (WWW) has profoundly altered the way we work, talk, relax, play, connect, affiliate, love, persuade, shop, sell, inform...in short, how we conduct the business of life. As end users, we perceive only the end to end flow of information. However, the network service providers have to constantly manage and monitor their networks in order to deliver seamless end-to-end services to their users. Network state models are tools used by the service providers for carrying out these tasks. In this dissertation, we propose two network state models for large scale Quality of Service (QoS) aware multi-class networks---one for performance analysis, and the other for state aggregation in inter-domain QoS routing.;For performance analysis, we have modeled the network as a finite shared resource, more specifically, as a multi-dimensional loss system with bursty call arrival processes. We have solved for the approximate steady state distribution of the above system. This has been an open problem since 1981. It is well known that the actual call arrival processes in the network are indeed bursty in nature. Hence this model aims to capture the operating characteristics of the network under more general traffic load conditions.;For state aggregation in QoS routing, we have developed schemes which use local statistics at the routers to compute compressed network state models. These models are periodically advertised to the neighboring domains, and are used by them for their path computation decisions. The proposed scheme captures the time varying behavior of network resources, and the routing capacity between pairs of border routers, which helps in better routing performance. We also develop a new performance measure, called aggregate accuracy index, which tests the efficacy of an aggregation scheme.