Bandwidth allocation algorithm for supporting differentiated quality of services in the internet

Abstract

We consider a bandwidth optimization problem in a Generalized Processor Sharing (GPS) server with multiple class queues to minimize the total bandwidth such that QoS requirements for each class queue are satisfied. Since it is difficult to obtain analytically evaluate the performance such as the delay distribution for self-similar input traffic in a GPS server, an optimized bandwidth vector is obtained using simulation based optimization technique. The conventional optimization algorithm requires rather long simulation time to solve the problem by using exterior penalty function methods. We propose a weight vector adjustment scheme that improves the performance of some classes and degrades other classes. We propose a new bandwidth optimization algorithm based on the weight vector adjustment scheme without complex calculations. In numerical results, the required time to find the optimal bandwidth allocation in GPS servers is greatly reduced. Then, we characterize the performance of a GPS server. We show that we can find the unique optimal solution by using the proposed algorithm. In addition to the bandwidth optimization in the GPS server, an end-to-end bandwidth optimization algorithm is proposed. To support the end-to-end QoS requirements, a QoS partitioning algorithm is proposed to reduce the complexity of the end-to-end bandwidth algorithm. We use a linear model to obtain optimal parameters in GPS network. By partitioning the end-to-end QoS requirements into multiple single node QoS requirements, we can use a single node optimization algorithm to solve the end-to-end bandwidth optimization problem.