Intelligent eviction strategy for efficient flow table management in OpenFlow Switches

Abstract

An OpenFlow Switch (OF-Switch) stores the flow entries in a flow table, having confined capacity. The flow table is located in an extremely efficient associative memory called TCAM (Ternary Content Addressable Memory). Typically, a data center with 100 edge switches can generate up to eight million flows per second, whereas, an OpenFlow switch can save around 64K flow entries. A table-miss, generally, causes an exchange of messages between an OF-Switch and the controller. This switch-controller communication (to install flow rule for every new flow) incurs severe overhead. Furthermore, if the flow table is full, then the controller-driven purging of flow entries induces substantial latency. Our objective is to address these critical challenges. In this paper, we propose an efficient flow table management proposition through intelligent autonomous (within OF-Switch) eviction mechanism. Instead of relying entirely on the expiry period alone of a flow entry, our eviction strategy involves smart data logging using highly space-efficient data structure - Multiple Bloom Filters (MBF) to determine candidate flow entries to be purged. The MBF, located in Static RAM (SRAM), is designed in a Column-major order. It constructs the flow's importance based on reference locality and recentness; using simple and near optimal collision-free hash function. We have performed flow logging using a real packet trace, with an error probability of less than 1%. The simulation results show around 37% improvement (on average) in the table-hit ratio compared to Least Recently Used (LRU) method in 2K-size flow table.