Declarative scheduling for optimally graceful QoS degradation

Abstract

The paper introduces a mechanism for optimally graceful QoS degradation in distributed real time multimedia services, declarative scheduling. We assume that for each medium of a multimedia service there are a set of implementations, both with differing resource consumption and provided QoS, thus allowing to exploit both resource-resource and resource-QoS tradeoffs. If the amount of available resources changes, the service may locally adapt without changing the provided QoS by scheduling a different set of implementations with different resource requirements. Should this kind of rescheduling not be possible, the service may adapt by scheduling implementations with a lower QoS providable with the resources available. In the latter case declarative scheduling degrades the QoS optimally, i.e. the difference between the provided QoS and the requested QoS is minimal with respect to the user's preferences. The problem, which implementations to select to minimize the QoS degradation is shown to be a so far unconsidered instance of the NP complete knapsack problem. A heuristic for this knapsack problem is proposed and shown to be sufficiently efficient and accurate to implement declarative scheduling.