TCAM-based packet classification for many-field rules of SDNs

Abstract

Software-defined networking (SDN) provides an emerging paradigm for future network architectures. OpenFlow is a widely deployed south-bound protocol for SDN. It uses rule-based packet forwarding and each rule may support many header fields for fulfilling various SDN applications. As the number of inspected header fields increases, these long rules may complicate the procedure of packet classification based on TCAM because TCAM only supports limited word sizes. In this paper, we propose a TCAM-based packet classification scheme to support many-field rules. Our scheme incorporates multiple stages of TCAM accesses, where each stage only compares selected bits of rules. Although our scheme may result in additional TCAM accesses, it enables packet classification upon rules longer than TCAM words. The experimental results show that our scheme can search 10K 247-bit rules by using 72-bit TCAM entries, where each packet classification can be accomplished with less than ten TCAM accesses. It also reduces the required TCAM storage by a magnitude of order. In summary, with our scheme, both feasibility and scalability of TCAM-based packet classification for SDN rules are significantly improved.