On precision and scalability of elephant flow detection in data center with SDN
Conghui Bi, Xuan Luo, Tong Ye, Yaohui Jin · 2013
Software Defined Network (SDN) provides better network management and higher utilization for data center network, which decouples control plane from data plane. The centralized control of entire network may result in large overhead and limit the scalability of control plane. Previously, some researches [6] [7] proposed an approach for reducing work load on controller with elephant flow detection. However, the threshold of the detection system was pre-configured with a fixed value without consideration of the dynamically changing traffic characteristics in data center, which could cause high detection error rates. In this paper, we propose a two-stage adaptive elephant flow detection system that could precisely identify elephant flows with low-overhead on controller. The main challenge is to configure the relevant decision threshold adaptively. We formulate it as an optimization problem and present an optimal solution. Analysis and experiment showed the benefits of our system over previous methods.