1. Towards TCAM-based Scalable Virtual Routers
2019/10/28
Towards TCAM-based Scalable Virtual Routers
Author: Layong Luo, Gaogang Xie, Steve Uhlig, Laurent Mathy, Kavé Salamatian, and Yingke Xie
Publisher: ACM CoNEXT 2012
Presenter: Yi-Hao Lai
Date: 2015/9/23
Department of Computer Science and Information Engineering
National Cheng Kung University, Taiwan R.O.C.
CSIE CIAL Lab 1

2. Introdution
Network virtualization allows multiple virtual networks to coexist on an underlying shared substrate, in isolation from each other.
When multiple virtual router instances are supported on the same physical equipment, we will use the term virtual router platform.
National Cheng Kung University CSIE Computer & Internet Architecture Lab

3. Virtual router platform
Multiple virtual router instances coexist and each has its own FIB (Forwarding Information Base).
FIB completion: merges the prefixes from all the FIBs into one TCAM.
FIB splitting: mixes the advantages of merged and non-shared data structures by splitting the prefixes into two prefix sets.
National Cheng Kung University CSIE Computer & Internet Architecture Lab

4. non-shared Virtual router ID (VID)
National Cheng Kung University CSIE Computer & Internet Architecture Lab

5. merged
If two entries from different FIBs share the same prefix, they can be merged into a single entry in the TCAM-based lookup engines.
NH array
National Cheng Kung University CSIE Computer & Internet Architecture Lab

6. merged Union Prefix Set (UPS)
National Cheng Kung University CSIE Computer & Internet Architecture Lab

7. FIB completion merged trie A B
2019/10/28
FIB completion
merged trie
Masking A B
National Cheng Kung University CSIE Computer & Internet Architecture Lab
CSIE CIAL Lab

8. FIB completion incoming packet prefix: 011, VID: 1 Result: B1
2019/10/28
FIB completion
Original NH
Completed NH
incoming packet
prefix: 011, VID: 1
Result: B1
National Cheng Kung University CSIE Computer & Internet Architecture Lab
CSIE CIAL Lab

9. FIB completion
insert
National Cheng Kung University CSIE Computer & Internet Architecture Lab

10. FIB completion
delete
National Cheng Kung University CSIE Computer & Internet Architecture Lab

11. FIB completion
masking prefix correction: When an update occurs in a node of the trie, some completed NHs in the sub-trie rooted at that node will have to be updated to reflect correct matching.
National Cheng Kung University CSIE Computer & Internet Architecture Lab

12. FIB splitting
If only disjoint prefixes are merged in the TCAM, the prefix masking issue does not exist any longer, as at most one prefix in a disjoint prefix set can match a given IP address.
About 90% of prefixes in the tries built from real FIBs are leaf prefixes.
National Cheng Kung University CSIE Computer & Internet Architecture Lab

13. FIB splitting
The disjoint leaf prefixes in the merged trie are stored in one TCAM based on the basic merged data structure.
The remaining prefixes in the merged trie are stored in another TCAM using the non-shared approach.
National Cheng Kung University CSIE Computer & Internet Architecture Lab

14. FIB splitting
National Cheng Kung University CSIE Computer & Internet Architecture Lab

15. FIB splitting
National Cheng Kung University CSIE Computer & Internet Architecture Lab

16. FIB splitting
update
National Cheng Kung University CSIE Computer & Internet Architecture Lab

17. FIB splitting
update
National Cheng Kung University CSIE Computer & Internet Architecture Lab

18. TCAM size
For IPv4 FIBs, the largest prefix length is 32 bits. As we can only get a maximum of 14 IPv4 FIBs for the evaluation, a 4-bit VID is enough to identify them.
For IPv6 FIBs, the longest prefix is 128 bits. Given a 4-bit VID, the longest virtual prefix is 132 bits.
National Cheng Kung University CSIE Computer & Internet Architecture Lab

19. TCAM size
For 14 full IPv4 FIBs, the proposed approaches can achieve a TCAM memory reduction of 92% and 82%
National Cheng Kung University CSIE Computer & Internet Architecture Lab

20. Total Cost of the System
National Cheng Kung University CSIE Computer & Internet Architecture Lab

21. Lookup and Update Performance
2019/10/28
Lookup and Update Performance
The worst-case update overhead in FIB completion is 2,206 write accesses per update
National Cheng Kung University CSIE Computer & Internet Architecture Lab
CSIE CIAL Lab

22. CCDF of update overhead
National Cheng Kung University CSIE Computer & Internet Architecture Lab