TCAM-Based Classification Using Divide-and-Conquer for Range Expansion

Ternary Content-Addressable Memory (TCAM) is the de facto industrial standard to perform packet classification. However inefficient representation of port ranges results in the range expansion problem which sharply degrades TCAM storage performance. A range has to be converted into a set of prefixes with each stored in a separate TCAM entry. The range expansion problem occurs when a rule with multiple range fields causes a multiplicative expansion in the number of TCAM entries. Unfortunately, the problem is growing worse as an increasing number of such rules in "real-world" classifiers are in use. To address range expansion our Divide-and-Conquer Scheme (DCS) fulfills the Divide-and-Conquer principle in two levels. First, we divide an individual range through range partitioning. A class of ranges can be optimally represented through a novel range encoding we developed. We observe the extensive presence of DCS-compatible ranges in real classifiers and more can be retrieved through our partitioning scheme. Second we divide the ranges in a classifier in terms of a hybrid utilization of various schemes. Technology advancement provides the necessary support for an open and flexible logical TCAM block division in order to avoid expensive hardware modifications and allow the use of DCS directly upon TCAM blocks. Our scheme allows fast preprocessing, constant time searching, and dynamic incremental update. Theoretical analysis and simulation using synthetic classifiers show a substantial storage improvement using our scheme.