Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Many IP forwarding engines have used TCAM(Ternary Content Addressable Memory) for IP address lookup. While TCAM has the advantage of parallel lookup in a single step, it has a difficulty in updating a prefix table. TCAM requires the prefix table to be ordered for the longest prefix matching, so each update may incur several memory movements to maintain the ordering. For the fast updating of the prefix table, it is needed to eliminate the ordering constraint. If a prefix table is partitioned into several sets each of which contains disjoint prefixes, there is at most one match in each set and the ordering of prefixes is not needed any more for the longest prefix matching. Prefixes of each partitioned table should be kept disjoint after updating as well as in the initial configuration. In this paper we present several algorithms for scheduling of updates to keep prefixes of partitioned tables disjoint. We compare and evaluate our scheduling algorithms by simulation with a real-world routing table and updates. Experiment results show that most of updates can be scheduled to the partitioned tables having disjoint prefixes so good update performance can be achieved.