Algorithms design for IPv6 address lookup and packet classification

Abstract

In addition to the conventional 1-D address lookup used to determine the next-hop of an incoming packet, an IP router is also required to perform multi-field packet classification in order to support advanced functions, such as network security, quality of service, and virtual private network etc. Many algorithmic and Ternary Content Addressable Memory (TCAM) based approaches have been published for the address lookup and packet classification problems. In general, software implementations are suitable for low-end routers, whereas hardware search engines are employed in high-end routers that demand a very high packet-processing rate. Most of the existing algorithmic methods for address lookup exploit the characteristics of IPv4 routing tables to speed up the processing rate. The adoption of IPv6 with 128-bit addresses poses great challenges to the existing methods. In this research, we investigate the necessary modifications and extensions to several well-known address lookup algorithms for application to IPv6. It is found that a hybrid method that combines level-compressed trie with hashing will give the best performance. Multi-field packet classification is a much harder problem compares to the 1-D address lookup. TCAM-based approach is the dominant solution used in the industry. With the adoption of IPv6, the cost of TCAM-based search engine is jet up while the efficiency is degraded. In this research, we develop a novel encoding scheme, called prefix inclusion coding (PIC), which can compress a 5-tuple 296-bit IPv6 filter to 72 bits (or even 36 bits for classifiers with up to a few thousands filters). Hence, the overall TCAM storage requirement can be reduced by 90%.