Stochastic Partitioning Strategies for Providing Scalable of Quality of Service Routing in Large Networks

Abstract

In today's Internet with its multiple services, there is a growing need to provide quality of service (QoS) guarantees tailored to specific services rather than the best-effort QoS that is currently provided. One of the most important mechanisms for providing QoS guarantees is QoS routing. However, the main drawback to the use of QoS routing algorithms is their inability to scale well to large networks. This is particularly problematic in the Internet where the number of nodes (switches, routers, terminals) is growing exponentially with time. One solution to deal with this scalability problem is to divide a network into manageable blocks (sub-systems) and this is known as network partitioning. This paper focuses on a review of the so called stochastic partitioning methods that dynamically change a network partition according to the traffic patterns in the network in order to minimise an objective function that reflects the computational effort involved in the underlying routing algorithms used in the network. Such methods therefore exploit the interactions between the network topology, the traffic flows and the routing which are not collectively considered by other forms of partitioning. The review suggests that a method called Clique Stochastic Partitioning looks to be the most promising and this is examined further in the paper. A recommendation of the review is that future large networks, such as the Internet, should be designed to exhibit completely connected sub-structures in their topologies (called cliques). Then stochastic partitioning can be used to scale any routing algorithms so that the computational effort remains bounded to a manageable value as the number of network nodes grows.