Fast updating algorithm for TCAMs using prefix distribution prediction

Abstract

Ternary Content Addressable Memories (TCAMs) are widely used in routers for storing and searching IP addresses. With ever-more devices being connected to the internet, TCAMs have become the de-facto means for performing address lookups. Prefixes have to be maintained in the descending order of prefix lengths in the TCAM. When a new prefix is inserted, the existing prefixes in the TCAM may have to be shuffled to retain the sorted order. Prefix shuffles decrease the update rate significantly. In this paper, we propose an algorithm for addressing the important issue of improving the update rate of prefixes in TCAMs. The main idea behind our algorithm is to use prefix length distribution prediction for reducing the number of prefix shuffles and consequently improve the update rate. The distribution of the prefix lengths of the first N prefixes is then computed. The inserted prefixes are then rearranged in the TCAM by inserting holes into the TCAM based of the computed prefix length distribution. The remaining prefixes are then inserted into TCAM. Since the distribution of the remaining prefixes matches closely with the predicted prefix distribution in real life traces, the number of shuffles reduces and significantly improves the update rate. Our algorithm has been evaluated on real-life traces of routers. Our findings show clearly that on real-life traces the prefix distribution prediction is fairly accurate. We have compared our algorithm with the algorithm which does not insert holes, the algorithm which uniformly distributes holes and the algorithm which maintains holes in the centre of the TCAM. The results show that our algorithm significantly reduces the number of shuffles and out performs the other algorithms. The algorithm can be implemented in software in the control plane and does not require any modification to the TCAM hardware making it easier to be rolled out into systems that already use TCAMs.