20
Authors: Wencheng Lu and Sartaj Sahni, Fellow, IEEE Conf. : IEEE/ACM Transactions on networking, 2010 Presenter : JHAO-YAN JIAN Date : 2011/3/15 Low-Power TCAMs for Very Large Forwarding Tables 1

Low-Power TCAMs for Very Large Forwarding Tables

Embed Size (px)

DESCRIPTION

Low-Power TCAMs for Very Large Forwarding Tables. Authors: Wencheng Lu and Sartaj Sahni , Fellow, IEEE Conf. : IEEE/ACM Transactions on networking, 2010 Presenter : JHAO-YAN JIAN Date : 2011/3/15. The simplest TCAM solution. - PowerPoint PPT Presentation

Citation preview

Page 1: Low-Power TCAMs for Very Large Forwarding Tables

Authors: Wencheng Lu and Sartaj Sahni, Fellow, IEEE

Conf. : IEEE/ACM Transactions on networking, 2010Presenter : JHAO-YAN JIANDate : 2011/3/15

Low-Power TCAMs for Very Large Forwarding Tables

1

Page 2: Low-Power TCAMs for Very Large Forwarding Tables

The simplest TCAM solutionThe simplest TCAM solution to packet

forwarding requires one TCAM search and one SRAM access to forward a packet.

The power consumption is that for a TCAM of this size.

2

Page 3: Low-Power TCAMs for Very Large Forwarding Tables

SUBTREE SPLIT(CoolCAM)

3

Page 4: Low-Power TCAMs for Very Large Forwarding Tables

POSTORDER SPLIT(CoolCAM)

4

Page 5: Low-Power TCAMs for Very Large Forwarding Tables

SUBTREE SPLIT

5

Page 6: Low-Power TCAMs for Very Large Forwarding Tables

SUBTREE SPLIT

6

Page 7: Low-Power TCAMs for Very Large Forwarding Tables

POSTORDER SPLITThe DTCAM buckets are filled using as

many rounds of feasible tree carving and packing as needed to completely carve out .

In each round, we select the bucket B with the fewest forwarding-table prefixes.

The preorder traversal.

7

Page 8: Low-Power TCAMs for Very Large Forwarding Tables

POSTORDER SPLIT(SP1)

8

Page 9: Low-Power TCAMs for Very Large Forwarding Tables

POSTORDER SPLIT(SP2)

9

Page 10: Low-Power TCAMs for Very Large Forwarding Tables

SIMPLE TCAM WITH WIDE SRAM

10

Page 11: Low-Power TCAMs for Very Large Forwarding Tables

SIMPLE TCAM WITH WIDE SRAM25 (50) is a rather crude estimate of the number

of instructions needed to process a 72-bit (144-bit) suffix node.

We can make 100 M SRAM accesses per second using a 10-ns SRAM and perform about 20 G instruction per second using a 3.2-GHz dual-core processor.

Therefore, about 200 instructions can be executed in the time it takes to make an SRAM access.

The time to process a 72-bit (144-bit) suffix node is only about 1/8 (1/4) that of an SRAM access.

11

Page 12: Low-Power TCAMs for Very Large Forwarding Tables

12

Page 13: Low-Power TCAMs for Very Large Forwarding Tables

13

Page 14: Low-Power TCAMs for Very Large Forwarding Tables

14

Page 15: Low-Power TCAMs for Very Large Forwarding Tables

EXPERIMENTAL RESULTS(1)

15

Page 16: Low-Power TCAMs for Very Large Forwarding Tables

EXPERIMENTAL RESULTS(2)

16

Page 17: Low-Power TCAMs for Very Large Forwarding Tables

EXPERIMENTAL RESULTS(3)

17

Page 18: Low-Power TCAMs for Very Large Forwarding Tables

EXPERIMENTAL RESULTS(4)

18

Page 19: Low-Power TCAMs for Very Large Forwarding Tables

EXPERIMENTAL RESULTS(5)

19

Page 20: Low-Power TCAMs for Very Large Forwarding Tables

20