A low-latency two-tier measurement and control platform for commodity SDN

Abstract

SDN is a centrally managed network architecture that automatically and dynamically allocates network resources to different types of workloads according to their application characteristics as well as current network conditions. We note that state-of-the-art SDN measurement mechanisms generally are slow, require reserved network resources, or use protocol-specific information when collecting flow-level statistics and constructing global network views. The lack of a fast and accurate SDN measurement mechanism significantly limits the effectiveness of SDN. In this article, we focus on the design of a scalable, accurate, and fast SDN measurement mechanism. We exploit packet-level statistics and propose a low-latency load-aware two-tier measurement and control platform for SDN. The port counter of the SDN switch, which tracks the number of bytes and packets that are sent and received on each port, is used for estimating link utilizations. The SDN switch actively monitors link states and pushes the statistics back to the SDN controller periodically for the detection of network congestion. When necessary, the controller can retrieve the flow statistics of the congested link from the SDN switches for traffic engineering. We implement a small-scale testbed and conduct emulations for evaluation. The experiment results show that compared to OpenSample and Planck, our two-tier mechanism quickly and accurately estimates link utilization and controls flow routes for better bisection throughput.