Efficient flow detection and scheduling for SDN-based big data centers

Abstract

In Software defined networking (SDN) based datacenters, flow-level management seriously limits system scalability due to large amount of control messages between data and control planes; and mice flows often are blocked by elephant flows because of the indiscriminate flow scheduling. To improve management efficiency and system performance, it is prerequisite to schedule elephant and mice flows respectively. Unfortunately, existing flow scheduling approaches in SDN consider only elephant flows. In this paper, we firstly propose an efficient flow detection mechanism. Then, we propose a novel DIFFERENtiated sChEduling (DIFFERENCE) approach that dynamically sets up paths for elephant and mice flows separately, based on current link workload. Our DIFFERENCE schedules mice flows with proactively installed weighted multipath routing algorithm and adjusts path weight according to link utilization. Instead, we propose a blocking island based path setup algorithm for elephant flows, which find the least congested path with shorter searching space. To balance traffic in a SDN networks, we design an algorithm to dynamically reschedule data flows in terms of current link utilization ratio. Experiment results on real public datacenter traces demonstrate that our approach outperforms related proposals in terms of various system performance.