A Scalable Per-flow Priority Scheduling Scheme for High-Speed Network

Abstract

In order to provide the service differentiation for various network applications, and guarantee delay and bandwidth requirement, packet scheduling is considered as a hot research topic and a crucial module in network device. In high speed network, it is hard to maintain and schedule a great number of queues for millions of in-progress flows in memory in line speed. In this paper, we propose a scalable per-flow scheduling scheme using a small fast memory to achieve fine-grained service guarantee. A limited number of queues are dynamically shared among concurrent flows based on the interesting fact that the number of simultaneous active flows is only in hundreds whatever the link speed is. The scheduling scheme is in a scalable hierarchical manner, in which the first layer supplies service differentiation and the second guarantees bandwidth and delay. We also implement an instance based on this scheme called DQS-SPQ-DRR (Dynamic Queue Sharing-Strict Priority Queue-Deficit Round Robin). Experiments based on real and synthetic traces are conducted to evaluate the DQS-SPQ-DRR. The results demonstrate that DQS-SPQ-DRR is well held in small memory and supplies per-flow service guarantee.