A memory efficient algorithm for packet classification
WASA'06 Proceedings of the First international conference on Wireless Algorithms, Systems, and Applications
| Hi-index | 0.00 |
The traffic on the Internet has exploded in recent years, due to a variety of new applications. Increasing the speed of packet transmissions and packet processing is a must. Optical fibers have already been used to make the transmission speed much faster, and there is still the potential to increase the bandwidth even wider in the near future. Routers as packet processing devices have to keep up with the pace of transmission speed. Looking for the next hop of the packet (IP address lookup) and differentiating the traffic according to some criteria (packet classification) are among the bases in packet processing. The common feature of IP address lookup and packet classification is that both of them need to consult a table to determine the best matching rule. Table lookups for the next generation IP networks becomes a promising topic. This thesis is focused on addressing solutions to such problems. The goal for developing the algorithms is to optimize the search strategy to meet the requirements of the next generation IP networks. In that, the key issues are efficient memory access time and storage consumption, easy implementation, and scalability to different tables and IPv6. In this thesis, one IP address lookup algorithm, one packet classification algorithm, and efficient TCAM applications are proposed. An IP address lookup algorithm based on Comb Extraction Scheme is proposed in Chapter 4. The target is to transform a sequential IP address process to a pair of smaller balanced sub-lookup processes and a comparison process, which are able to perform in parallel. The experimental results show that it has outstanding performance in both average search speed and storage cost. A packet classification algorithm using bit characteristics is proposed in Chapter 5. The target is to optimize the search path for large classifiers. This algorithm consists of SMS-B three stages, hash table lookup, optimal search trie lookup, and small set lookup. The search path decision needs to take into account the bits chosen from all the dimensions at each step. It can overcome the problem of an unbalanced distribution of rules among dimensions. The experimental results provide strong support for the algorithm design. Techniques that allow TCAMs more efficient in table lookups are proposed in Chapter 6. Three techniques are proposed to reduce TCAMs power consumption, increase the capability to handle bigger tables, and solve range matching problems in TCAMs, respectively. The experimental results verify that they have excellent performances. The prominent feature of our algorithms is that they use novel optimal lookup methods, avoiding the traditional search order. They are amenable to parallel implementations. The experimental results show that they are not only comparable to current solutions but also have better performance for the table lookup at the edge. We describe the background in this field and review some existing classical technologies of table lookups in the first three chapters. After that, the proposed algorithms are detailed in each of the following three chapters. Finally, we conclude in the last chapter.