On the Optimization of Flow Tables of SDN-enabled Switches

In SDN network, the fine grained flow matching mechanism makes number of flow entries extremely huge so that flow tables have become the bottleneck of large-scale deployment of SDN. To solve this problem, one of the known solutions is to make a hybrid flow table structure that utilizes both TCAM and other memory, such as SRAM. The motivation is to store flow entries with many wildcards or frequently used in TCAM and to store other flow entries especially those need fine grained matching in other memory to get a cheaper, high capacity flow table with acceptable matching speed. This paper follows the hybrid flow table architecture and focuses mainly on how to make high capacity and accepted matching speed flow tables. We use a Bloom Filter pipeline to store flow entries, design and implement an intermediate adaptation layer between controllers and Openflow switches to transform the corresponding messages. The experiments results show that the flow table space can be effectively reduced to 33 bits for one flow entry, and the packet matching time can be substantially reduced compared with CPqD software switch.