3 Chapter - Efficient packet classification with digest caches

Abstract

Packet classification allows in-network devices such as firewalls, network address translators, and edge routers to provide differentiated service and access to network and host resources by efficiently determining how the packet should be processed. These services require a packet to be classified so that a set of rules can be applied to such network header information as the destination address, flow identifier, port number, or layer-4 protocol type. Typical packet classification caches trade off size and performance. This chapter proposes a novel cache architecture that efficiently and effectively uses memory, given a slightly relaxed accuracy requirement. Performance of any existing flow-caching solution that employs exact caching algorithms can be improved dramatically by employing this technique, at the sacrifice of a small amount of accuracy. This new technique is superior to the previous Bloom filter approximate caching algorithms, in both theoretical and practical performance, while also addressing the shortcomings in the Bloom filter cache design without introducing any additional drawbacks. This technique can be applied to the design of a novel two-level exact cache, which can take the advantage of hierarchical memory to accelerate exact caching algorithms, with strong results.