| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系所 === 101 === Ternary content addressable memory (TCAM) has been used to perform packet classification in network devices due to its superior speed performance. However, TCAM has several drawbacks in supporting multi-match packet classification. In this paper, we present a TCAM-based scheme for multi-match packet classification. Our scheme partitions a rule database based on layer information for database-dependent range encoding. Our idea is based on an observation that range encoding schemes usually minimize the number of layers. The rule partitioning generates rule groups which satisfy that the rules in a group do not overlap with each other. Each rule is then tagged a bitwise group identification to fulfill multi-match packet classification. Two approaches, loose coupling and tight coupling, are derived with difference independence between search procedure and range encoding. Both approaches have different performance tradeoff. Accordingly, two refinements are presented. The first refinement uses all available TCAM entries to shorten relative long bitmaps caused by the first approach. The second refinement eliminates extra TCAM accesses of the second approach by using pre-computation. In particular, combining loose coupling approach with the method using all available TCAM entries can effectively exploit TCAM space to achieve fewer TCAM accesses than the number of matching rules. The experimental results show that combining range encoding with multi-match packet classification has advantages of storage efficiency and speed superiority.
|
|---|
