【摘要】 Internet区分服务(DiffServ)中EF PHB(Expedited ForwardingPer Hop Behavior)提供严格的端到端延迟保证,其实现机制和性能是当前研究的热点。随着可扩展性成为核心网络考虑的关键因素,一般用简单的FIFO调度实现EF PHB。FIFO实现问题在于最坏的端到端延迟与流经历的最大跳数成正比,结果降低了网络最坏延迟性能,并影响了整个网络的总体利用率。文章在分析并比较FIFO实现以及考虑流跳数因素的绝对跳数优先(HBAP)实现、相对跳数优先(HBRP)实现的延迟性能基础上,提出了用基于剩余路径跳数的动态优先(DHBP)调度实现EF PHB。理论分析和实验结果表明,基于剩余路径跳数的动态优先调度算法可以平衡不同跳数流的端到端延迟性能,从而减小网络最坏的端到端延迟,并有效地提高了网络的利用率,最坏延迟性能明显优于FIFO和绝对跳数优先调度,与性能最优的相对跳数优先调度相似,并将计算复杂度降为O(1)。更多还原

【Abstract】 As the scalability of implementation of Expedited Forwarding Per Hop Behavior (EF PHB) becomes a crucial problem in Diffserv Framework, the simple FIFO packets scheduling algorithms attracts more attentions recently. In our study, the Hop counts Based Relative or Absolute Priority scheduling (HBRP/HBAP) algorithms have been proposed to balance the differences of end-to-end delay of flows that traverse paths of different hop counts, which decrease the worst end-to-end delay bound of whole domain and improve the maximum allowable network utilization level accordingly. However, the computational complexity of the HBRP algorithm is in the order of the network diameter. In this paper, we propose a new Dynamic Hop counts Based Priority scheduling algorithms (DHBP) with complexity of O(1), which has similar performance to the HBRP. We compare the performance of the DHBP with those of FIFO, HBAP and HBRP algorithms by theoretic analyses and simulation experiments.更多还原