Class-based Flow Scheduling Framework in SDN-based Data Center Networks

Abstract

The emerging technologies leveraging Data Center Networks (DCN) and their consequent traffic patterns impose more necessity for improving Quality of Service (QoS). In this paper, we propose Sieve, a new distributed SDN framework that efficiently schedules flows based on the available bandwidth to improve Flow Completion Time (FCT) of mice flows. In addition, we propose a lightweight sampling mechanism to sample a portion of flows. In particular, Sieve schedules the sampled flows, and it reschedules only elephant flows upon threshold hits. Furthermore, our framework allocates a portion of the flows to ECMP, so that the associated overhead can be mitigated in the control plane and ECMP-related packet collisions are fewer as well. Mininet has been used to evaluate the proposed solution, and Sieve provides better FCT up to 50% in comparison to the existing solutions like ECMP and Hedera.