Allocating commodity resources in aggregate traffic networks

Abstract

We examine the relation between cost and quality in networks which carry aggregate traffic. A powerful tool in this is the burstiness or indifference curve associated with a stochastic traffic flow. We relate burstiness to quality and use this relation to explore the quality experienced by aggregated flows under various rules for allocating resources to them. An example is motivated by the controlled load service specification. We show how the imposition of costs associated with buffer space and service capacity leads to the notion of a cost-optimal allocation of resources. This defines the cheapest operating point in a network where resources are commodities to be purchased as necessary to satisfy quality requirements. We define a notion of cost-based admission control: a linear admission rule which can be based on declared or measured traffic parameters.