Performance evaluation of packet processing architectures using multiclass queuing networks

Abstract

Modern real-time systems consist of complex parallel and heterogeneous architectures. Early design decisions, such as the partitioning of functionality onto architecture building blocks and the choice of algorithms, have a large impact on the quality of the resulting platform. In order to support the designer during this concept phase, we have developed our performance evaluation framework SystemQ. In this paper, we demonstrate why multiclass queuing networks as used by SystemQ are a natural abstraction for evaluating network processing platforms. In particular, we reveal the impact of scheduling policies on the quality-of-service, such as the residence time of network traffic in the system. For the same stimuli, the packet latency can vary by more than one order of magnitude if only one queuing discipline in the system is modified.