Accelerating Packet Classification with Two Class Cuckoo Filters (TC-CF)

Abstract

Flexible and high speed packet classification is a key element to enable Software Defined Networks (SDNs) but implementing large matching tables with rules that contain wildcard bits at high speed is challenging. This paper considers the use of Cuckoo Filters (CF) to accelerate packet classification implemented with hash tables. In particular, the filters are used to reduce the number of hash table accesses needed for each packet lookup. In this context, it is beneficial to use a filter that can support two types of elements, popular and unpopular so that patterns with fewer elements (unpopular) get a lower false positive rate. The paper introduces a new structure, the Two Class Cuckoo Filter (TCCF) that efficiently supports two types of elements. The TCCF is evaluated and compared to the use of a plain CF or the existing LACF for several sets of rules. The results show that the TCCF can support both popular and unpopular elements regardless of the filter occupancy and outperforms the other alternatives. Therefore, it can be an interesting option to accelerate packet classification implemented with hash tables in SDN applications.