Performance analysis of coordinated network scheduling

[摘要] In multi-hop networks, packet schedulers at downstream nodes have an opportunity to make up for excessive latency at upstream nodes. Similarly, when packet experience good service, and low delay at upstream nodes, downstream nodes can reduce priority and schedule other packets first. A framework that provides such a property is called Coordinated Network Scheduling (CNS). In this thesis, we analyze the performance of CNS. First, we review the general CNS definition, key concepts and admission control condition. We then use simulations to experimentally evaluate the end-to-end performance properties of coordinated scheduling. For example we show that CNS is robust to parameter allocation as it achieves almost the same end-to-end performance for very different per-node delay budget allocation schemes. Next, we explore the accuracy of the theoretical admission control tests, and find that theoretical performance predictions are quite close to measured values. We then compare CNS with the alternate approach that uses smoothing with bufferless multiplexing and find that CNS is able to achieve significantly better performance. Finally, we show that CNS schedulers are able to limit traffic distortion to a narrow range resulting in improved end-to-end performance and more efficient resource utilization.

[发布机构] Rice University

[效力级别] Electrical engineering