MDCP: Measurement-Aware Distributed Controller Placement for Software Defined Networks

Abstract

The rapid development of software defined measurement has significantly improved network measurement and monitoring. The key challenge for software defined measurement is to design a low-cost measurement framework which has minimum impact on the network. The state-of-the-art approaches mainly focus on reducing the measurement overhead by sampling or aggregation. However, little attention has been devoted to eliminating this issue in the physical layer. We observe that the placement of the controllers significantly affects the measurement overhead for software defined measurement. Based on this observation, we rethink software defined measurement frameworks and propose a novel scheme to minimize the measurement overhead. Our approach is application-agnostic, cost-effective and robust to traffic dynamics. We formulate the measurement-aware distributed controller placement (MDCP) problem as a quadratic integer programming problem, which takes both the synchronization cost and the flow statistics collection cost into account. Due to its high computational complexity, we develop two novel algorithms to efficiently approximate near-optimal placements. In particular, we employ an algorithm with an approximation ratio of 1.61 to obtain the placement in the discrete approximation algorithm. We conduct experiments on over 240 real network topologies and the results demonstrate the effectiveness of MDCP. Trace-driven simulations verify that our proposal is robust to traffic dynamics and can reduce 40% of the measurement overhead on average.