Resource Allocation for Quality of Service Provision in Buffered Crossbar Switches With Traffic Aggregation

Abstract

A wide spectrum of emerging networking applications demand much stricter requirements on computer network quality of service (QoS). Packet switches play one of the most crucial roles in network QoS provision. The buffered crossbar switching architecture is promising to achieve high performance without a complex implementation and is expected to be one of the dominant switch structures for high-speed networks. In this paper we study the problem of resource allocation to aggregated traffic classes in buffered crossbar switches for QoS provision. We develop the model and techniques for determining the required amount of bandwidth and buffer space for aggregated traffic classes to provide delay performance guarantees and analyze the impact of traffic aggregation on resource utilization in buffered crossbar switches. We consider two traffic aggregation cases: aggregation of a set of flows with identical traffic profiles and aggregation of a set of flows with different traffic parameters. For both cases, we compare the amount of resources required by the aggregated traffic class and by the set of individual flows for achieving identical delay objectives. We find that traffic aggregation improves the utilization of both bandwidth and buffer space in buffered crossbar switches. Our analysis shows that the amount of resources saved by traffic aggregation is associated with multiple factors, including the delay objective, the arrival traffic parameters, and the number of flows aggregated into one class. The model and techniques we develop in this paper are general enough to be applied to buffered crossbar switches with a variety of scheduling algorithms.