Low Power TCAM Forwarding Engine for IP Packets

1. Low Power TCAM Forwarding Engine for IP Packets Authors: Alireza Mahini, Reza Berangi, Seyedeh Fatemeh and Hamidreza Mahini Presenter: Yi-Sheng, Lin (林意勝) Date: Aug. 12, 2008 Publisher/Conf. : Military Communications Conference, 2007. MILCOM 2007. IEEE Dept. of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C.

2. Outline • Introduction • Proposed Approach • Proposed Architecture • Lookup operation • Update Operation • Performance Evaluation • Conclusion

3. Introduction • The work’s main objective is a TCAM-based router architecture that consumes less power. • The existing approaches reduce power either by routing table compaction or selecting a portion of the TCAM. • The author’s reduces power by combining the two approaches. • MLET : MultiLevel Enabling Technique.

4. Proposed Approach(1/4) • Routing table minimization : 1. Elimination of overlapped prefixes. 2. Partitioning the result of previous step into PRTs (Partial Routing Tables). 3. Minimizing the PRTs using Espersso Minimization Units (EMU) in parallel. [11] R. Lysecky and F. Vahid, “On-chip logic minimization,” In Proceedings of the 40th conference on Design automation, ACM Press, 2003, pp 334–337.

5. Proposed Approach(2/4)

6. Proposed Approach(3/4) • Overlap elimination : 1. |Pa| < |Pb|. 2. Pa,i= Pb,ifor all 1 < i < |Pa|. 3. There is no prefix Pc such that |Pa| < |Pc| < |Pb|, and Pc,i= Pb,ifor all 1≤ i ≤ |Pc| . Example : Pa : 1 0 1 0 1 0 0 0 port : A Pb : 1 0 1 0 * port : A A A

7. Proposed Approach(4/4) • Route Table Partitioning and PRTs : The partitioning rule is very straightforward. It partitions the prefixes based on their corresponding output port. • EMU ( [11] ) : Prefix 10011100 10001100 100*1100

8. Proposed Architecture(1/2) 10 11 01 01 10 1* 11 00 1* 10 01 00 ** 10

9. Proposed Architecture(1/2) • The separator unit (SU) extracts the destination IP address of incoming packet. • After IP address extraction and splitting, the split parts of IP address is stored into the Data Registers (DRs). • ES signals are Enabling Signals of each stage in rows. • ML signals are Match Line signals which are placed in the output of the last stage determine which row is matched with destination IP address. • The LPM selector unit, dose the Longest Prefix Matchingselection.

10. Lookup operation • The time complexity of lookup operation is O(k). (K is number of stages)

11. Update Operation • Insert : 1. According to the output port of pi prefix which is qi,the start and end addresses of PRTqi got from Merger Unit. 2. the PRT entries with new pisent to the related EMU. 3. Finally rebuilding of the table and saving the new addresses must be done. • Delete : The algorithm for removing a prefix from the routing table is more complex because the problem of prefix covering.

12. Performance Evaluation(1/4) • EPS (Enabled bits Per Search) : power consumption in lookup operation directly depends on the number of enabled bits of TCAM table cells. • MEPS (Mean Enabled bits Per Search): m : number of lookup times

13. Performance Evaluation(2/4) • POF (Power Optimization Factor):

14. Performance Evaluation(3/4)

15. Performance Evaluation(4/4)

16. Conclusion • A power efficient TCAM based hardware architecture has been proposed. • The power consumption is found to be remarkably low to promise efficient TCAM design in the future