A Tale of Two (Flow) Tables: Demystifying Rule Caching in OpenFlow Switches

Abstract

Software Defined Networking (SDN) enables flexible flow control by deploying fine-grained rules in OpenFlow switches. Modern commodity switches usually use TCAM to store these rules and perform high-speed parallel lookups. Though efficient, the TCAM capacity is limited because TCAM is expensive in cost and power-hungry. The explosive growth in the number of rules has exacerbated the limitation of TCAM. There have been considerable efforts in implementing hybrid flow tables with both TCAM and RAM, where the high-speed TCAM is regarded as a cache to store the most popular rules and the cheap RAM is used to handle cache miss. The primary challenges for designing hybrid TCAM/RAM flow tables lie in how to improve cache hit rate and how to handle wildcard rule dependency when allocating rules between TCAM and RAM. In this paper, we present the design and evaluation of CuCa, a practical and efficient rule caching scheme for hybrid switches. Different from existing schemes, CuCa offers both offline and online algorithms for rule caching, corresponding to the proactive and reactive approaches to OpenFlow rule installation. By designing a two-stage-cache architecture in TCAM, CuCa can handle rule dependency efficiently and provide remarkable performance improvements. Simulation and real-world experiment results reveal that CuCa improves average TCAM hit rate by 38.7% compared to state-of-the-art schemes and by over 33% compared to the default caching algorithm of a commodity OpenFlow switch.