Presentation is loading. Please wait.

Presentation is loading. Please wait.

Bit Weaving: A Non-Prefix Approach to Compressing Packet Classifiers in TCAMs Author: Chad R. Meiners, Alex X. Liu, and Eric Torng Publisher: 2012 IEEE/ACM.

Similar presentations


Presentation on theme: "Bit Weaving: A Non-Prefix Approach to Compressing Packet Classifiers in TCAMs Author: Chad R. Meiners, Alex X. Liu, and Eric Torng Publisher: 2012 IEEE/ACM."— Presentation transcript:

1 Bit Weaving: A Non-Prefix Approach to Compressing Packet Classifiers in TCAMs Author: Chad R. Meiners, Alex X. Liu, and Eric Torng Publisher: 2012 IEEE/ACM Transactions on Networking Presenter: Ching Hsuan Shih Date: 2014/01/22

2 Outline I. Introduction II. Related Work III. Bit Swapping IV. Bit Merging V. Experimental Results

3 I. Introduction A. Background on TCAM-Based Packet Classification B. Motivation for TCAM-Based Classifier Compression

4 I. Introduction (Cont.) C. Limitations of Prior Art All prior TCAM-based classifier compression schemes ([8]-[13]) suffer from one fundamental limitation: They only produce prefix classifiers.

5 I. Introduction (Cont.) D. Our Bit Weaving Approach The basic idea of bit weaving is simple: Adjacent TCAM entreis that have same decision and have a hamming distance of one can be merged into one entry by replacing the bit in question with *.

6 I. Introduction (Cont.) Bit weaving: bit swapping  dynamic programming techniques [9], [13] to find a minimal TCAM table for each partition  bit merging Bit weaving applies two new techniques: Bit swapping: First cuts a rule list into a series of partitions. Within each partition, a single permutation is applied to each rule’s predicate to produce a reordered rule predicate, which forms a single prefix. Bit merging: Finds and merges mergeable rules from each partition.

7 II. Related Work A. Classifier Compression [8]-[11], [13], [14] Classifier compression converts a given packet classifier to another semantically equivalent packet classifier that requires fewer TCAM entries.

8 II. Related Work (Cont.) B. Range Encoding [1], [17], [18], [19]-[23] Range encoding schemes cope with range expansion by developing a new representation for important packets and intervals. C. Circuit and Hardware Modification [24], [25], [26]

9 III. Bit Swapping A. Prefix Bit Swapping Algorithm Prefix bit-swaps are useful for two main reasons. First, we can use algorithms [8], [9], [27] that can optimally minimize prefix rule lists. Second, prefix format facilitates the second key idea of bit weaving, bit merging.

10 III. Bit Swapping (Cont.) B. Minimal Cross-Free Classifier Partitioning Algorithm Definition III.2: We use P(T) to denote the set of prefix bit-swaps for ternary string list T. Definition III.3 (Cross Pattern): Two ternary strings t 1 and t 2 form a cross pattern if and only if { i | t 1 [i] = * } ≠ { j | t 2 [j] = * } ˄ { j | t 2 [j] = * } ≠ { i | t 1 [i] = * }. In such cases, we say that t 1 crosses t 2. Given a classifier C, if P(C) = ϕ, we cut C into partitions where each partition has no cross patterns and thus has a prefix bit-swap.

11 III. Bit Swapping (Cont.)

12 C. Partial-List Minimization Algorithm [27]

13 IV. Bit Merging

14 IV. Bit Merging (Cont.)

15 V. Experimental Results We use RL to denote a set of 25 real-world packet classifiers on which we performed experiments. We then split RL into two groups, RL a and, RL b where the expansion ratio of direct expansion is less then 2 in RL a and the expansion ratio of direct expansion is greater than 40 in RL b. We have no classifiers where the expansion ratio of direct expansion is between 2 and 40. It turns out |RL a |=12 and |RL b |=13.

16 V. Experimental Results (Cont.)


Download ppt "Bit Weaving: A Non-Prefix Approach to Compressing Packet Classifiers in TCAMs Author: Chad R. Meiners, Alex X. Liu, and Eric Torng Publisher: 2012 IEEE/ACM."

Similar presentations


Ads by Google