Presentation is loading. Please wait.

Presentation is loading. Please wait.

HybridCuts: A Scheme Combining Decomposition and Cutting for Packet Classification Authors : Wenjun Li, Xianfeng Li Publisher : 2013 IEEE 21st Annual Symposium.

Similar presentations


Presentation on theme: "HybridCuts: A Scheme Combining Decomposition and Cutting for Packet Classification Authors : Wenjun Li, Xianfeng Li Publisher : 2013 IEEE 21st Annual Symposium."— Presentation transcript:

1 HybridCuts: A Scheme Combining Decomposition and Cutting for Packet Classification Authors : Wenjun Li, Xianfeng Li Publisher : 2013 IEEE 21st Annual Symposium on High-Performance Interconnects Presenter : Kai-Hsun Li Date : 2015/3/25 Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C.

2 Introduction EffiCuts suffers from too many memory accesses and a large number of separated trees. Decomposition based schemes have poor storage scalability. In this paper, we propose HybridCuts, a combination of decomposition and decision-tree techniques that improves storage and performance simultaneously. National Cheng Kung University CSIE Computer & Internet Architecture Lab 2

3 Proposed Scheme(1/11) -HybridCuts - Decomposition-based Framework(1/4) National Cheng Kung University CSIE Computer & Internet Architecture Lab 3 Definitions: Given an N-dimensional rule R = (F 1, F 2, F 3, …, F N ), Len i represents the length of field F i, and a threshold value vector T = (T 1, T 2, T 3, …, T N ). We call F i is a small field if Len i ≦ T i. Then we define the following concepts for R: Big rule: ∀ i ∈ {1, 2, 3... N}, F i in R is a big field Small rule: ∃ i ∈ {1, 2, 3... N}, F i in R is a small field

4 Proposed Scheme(2/11) -HybridCuts - Decomposition-based Framework(2/4) National Cheng Kung University CSIE Computer & Internet Architecture Lab 4

5 Proposed Scheme(3/11) -HybridCuts - Decomposition-based Framework(3/4) National Cheng Kung University CSIE Computer & Internet Architecture Lab 5 We decompose a 5-tuple rule set into the following five subsets without duplicates among each other: Big-subset: SA, DA, SP and DP are all big field SA-subset: SA is a small field for each rule DA-subset: DA is a small field for each rule SP-subset: SP is a small field for each rule DP-subset: DP is a small field for each rule

6 Proposed Scheme(4/11) -HybridCuts - Decomposition-based Framework(4/4) National Cheng Kung University CSIE Computer & Internet Architecture Lab 6

7 Proposed Scheme(5/11) -HybridCuts - FiCuts(1/2) National Cheng Kung University CSIE Computer & Internet Architecture Lab 7 Simplicity: FiCuts conducts cuttings on the subset along a fixed dimension. Adaptivity: FiCuts can decide when to stop FiCuts and resort to other more effective cutting methods.

8 Proposed Scheme(6/11) -HybridCuts - FiCuts(2/2) National Cheng Kung University CSIE Computer & Internet Architecture Lab 8

9 Proposed Scheme(7/11) -HybridCut - Multi-dimensional Cutting National Cheng Kung University CSIE Computer & Internet Architecture Lab 9 With the shrinking of the search space, rule replications begin to rise with fine cuts. FiCuts is able to detect this problem and signals switching to a more appropriate multi-dimensional cutting

10 Proposed Scheme(8/11) -HybridCuts National Cheng Kung University CSIE Computer & Internet Architecture Lab 10 binth=2 spfac=2

11 Proposed Scheme(9/11) -HybridCuts National Cheng Kung University CSIE Computer & Internet Architecture Lab 11

12 Proposed Scheme(10/11) -HybridCuts National Cheng Kung University CSIE Computer & Internet Architecture Lab 12 binth=2 spfac=2

13 Proposed Scheme(11/11) -HybridCuts National Cheng Kung University CSIE Computer & Internet Architecture Lab 13 binth=2 spfac=2

14 Optimization(1/2) National Cheng Kung University CSIE Computer & Internet Architecture Lab 14 Most rules have at least one small IP address field (SA or DA). Based on this observation, we reduce the number of decision trees. For a 5-tuple rule set, we decompose it into three subsets: Big-subset: Both SA and DA are big field SA-subset: SA is a small field for each rule DA-subset: DA is a small field for each rule

15 Optimization(2/2) National Cheng Kung University CSIE Computer & Internet Architecture Lab 15

16 EXPERIMENTAL RESULTS(1/3) National Cheng Kung University CSIE Computer & Internet Architecture Lab 16

17 EXPERIMENTAL RESULTS(2/3) National Cheng Kung University CSIE Computer & Internet Architecture Lab 17 The sizes of subsets

18 EXPERIMENTAL RESULTS(3/3) National Cheng Kung University CSIE Computer & Internet Architecture Lab 18


Download ppt "HybridCuts: A Scheme Combining Decomposition and Cutting for Packet Classification Authors : Wenjun Li, Xianfeng Li Publisher : 2013 IEEE 21st Annual Symposium."

Similar presentations


Ads by Google