clue: achieving fast update over compressed table for parallel lookup with reduced dynamic...

19
Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang Zhang, Huichen Dai and Bin Liu Publisher: IEEE ICDCS, 2012 Presenter: Kai-Yang, Liu Date: 2013/3/13

Upload: joleen-hamilton

Post on 04-Jan-2016

219 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup withReduced Dynamic Redundancy

Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang Zhang, Huichen Dai and Bin LiuPublisher: IEEE ICDCS, 2012

Presenter: Kai-Yang, Liu

Date: 2013/3/13

Page 2: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

INTRODUCTION

•To achieve high performance, backbone routers must gracefully handle the three problems: routing table Compression, fast routing Lookup, and fast incremental UpdatE (CLUE).

•CLUE consists of three parts: a routing table compression algorithm, an improved parallel lookup mechanism, and a new fast incremental update mechanism.

2

Page 3: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Compression Algorithm• ONRTC compresses the routing table size to 70%

of its original size. • Prefix overlap is eliminated.

3

Page 4: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

ONRTC Algorithm

4

Page 5: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Partition Algorithm•In order to achieve parallel lookup, the

prefixes should be split into partitions firstly.

•Step 1: compute the partition size. Suppose the size of routing table is M and the partition count is n, then the size of each partition is M/n.

•Step 2: traverse the trie by inorder, then put every M/n prefixes to each bucket.

5

Page 6: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Improved Parallel Lookup Mechanism

6

Page 7: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

The DRed update process of CLPL’s mechanism

7

Page 8: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

The DRed update process of CLUE’s mechanism

8

Page 9: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

The Incremental Update Mechanism

• The whole update process is divided into three steps :

1) trie update; 2) TCAM update; 3) DRed update.

• Time to Fresh (TTF) is defined in this paper, including TTF1 (TTF-trie), TTF2 (TTF-TCAM), and TTF3 (TTF-DRed).

9

Page 10: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Experiments on Compression by ONRTC

10

Page 11: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Partition comparison among the three algorithms

11

Page 12: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

TTF1 comparison between CLPL and CLUE

12

Page 13: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

TTF2 comparison between CLPL and CLUE

13

Page 14: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

TTF3 comparison between CLPL and CLUE

14

Page 15: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

TTF1+TTF2+TT3 comparison between CLPL and CLUE

15

Page 16: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

WORKLOAD ON DIFFERENT PARTITIONS AND TCAM CHIPS.

16

Page 17: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Load balance of workload distribution by CLUE

• Each TCAM takes 4 clocks to process a packet, while a packet arrives per clock. The FIFO is set to 256 and redundancy size is set to 1024 prefixes.

17

Page 18: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Speedup factor comparison between CLPL and CLUE

18

Page 19: CLUE: Achieving Fast Update over Compressed Table for Parallel Lookup with Reduced Dynamic Redundancy Author: Tong Yang, Ruian Duan, Jianyuan Lu, Shenjiang

Hit rate comparison between CLPL and CLUE

19