Self-organized multi-class QoS provision for ABR traffic in ATM networks

This paper presents a distributed, market-based approach for providing multiple service classes with graded quality-of-service (QoS) characteristics in the context of supporting ABR traffic in ATM networks. We define a network model facilitating two types of network traffic, reserved and nonreserved, the latter servicing traffic sources with non-strict QoS requirements on a best-effort basis. Two goals characterize the present approach: transparency of control and adaptability to changing application demand. A prioritized cell scheduling policy is defined which induces service classes with a range of QoS properties. Applications, in the process of trying to maximize individual utility, compete for bandwidth tagged by service class labels via a market mechanism. Greediness, in general, does not lead to system-optimal allocation of network resources, and system optimality, Nash equilibria, and Pareto optimality need not coincide. The main feature of our system is its ability to adapt to changing aggregate application demand and its QoS characteristics in the absence of explicit control where transparency extends both to applications and the network. Simulation results are shown which depict the self-organizing properties of the system