SIRS: selective intersected rule search for packet classification

Abstract

The key function of packet switches in the Internet, and all other packet switching networks, is packet classification. Techniques proposed for packet classification to date have many important characteristics such as speed, low cost, and flexibility but do not scale well to the needs of very large core network switches. This paper proposes a new scalable classification mechanism by extending an existing scheme, called bitmap intersection. The proposed scheme operates effectively at sizes over 200 times that of classifiers supported by the original scheme without sacrificing classification or update speed. Operation of the new technique was confirmed by computer simulation using 128,000 rules in a classifier and is compared with results reported in the literature. The simulations show that the proposed scheme is superior to both the original bitmap scheme and another existing scheme, pruned tuple space search (a comparable scheme in the literature with the proposed scheme), and can operate at forwarding rates compatible with OC192c transmission rate with reasonable memory consumption. These findings confirm that the scheme scales well for large classifiers and speeds necessary for core routers.