Fast updating scheme of forwarding tables on TCAM

1
Fast updating scheme of forwarding tables on TCAM

• Weidong Wu Bingxin Shi Feng WangModeling,
  Analysis, and Simulation of Computer and
  Telecommunications Systems, 2004. (MASCOTS 2004).
  Proceedings. The IEEE Computer Society's 12th
  Annual International Symposium on , 4-8 Oct. 2004
  Pages522 - 527

2
Outline

• Introduction
• Related works
• L-algorithm
• PLO_OPT
• 1-algorithm
• CAO_OPT
• Our approach
• Experiment
• Conclusion

3
Introduction

• The need to keep a sorted list of prefixes in
  TCAM make updates slow.
• Because the prefixes are in order of decreasing
  prefix length, the prefixes that are longer than
  the new prefix are move down to make free space
  for the new prefix, in the worst case, all
  prefixes are moved.
• Because of instability of internet, these changes
  of prefixes in routing table can occur at the
  rate of approximately 100-1000 prefixes per
  second.
• The slow update is the performance bottleneck of
  TCAM.
• In the paper, we focus on the problem of making
  TCAM-based forwarding engines more update
  efficient.

4
The basic architecture of TCAM
5
Related works
6
L-algorithm

• There are N prefixes of a routing table in TCAM,
  which are stored in order of prefix length.
• If the free space is at end of TCAM, show Fig, in
  the worst case, it is necessary to move N
  prefixes for the addition of a new prefix.

7
PLO_OPT

• If the free space is in the middle of TCAM, show
  Fig, in the worst case, it is necessary to move
  N/2 prefixes for the addition of a new prefix

8
1-algorithm

• If there are the free spaces between the
  different length prefixes, show Fig, the new
  prefix can be insert into the free space, it is
  not necessary to move any prefixes.
• But there more free spaces to waste.

9
CAO_OPT

• All prefixes are stored in TCAM with the
  chain-ancestor ordering constrain.
• D is the length of the chain, show Fig.
• If the free space is in the middle of the chain,
  in the worst case, it is necessary to move D/2
  prefixes for addition of a new prefix.
• If the free space is at the end of the chain, in
  the worst case, it is necessary to move D prefix
  for addition of a new prefix.

10
Proposition I

• If a prefix P matches IP address D, we call D?P.
• There are two prefix P1, P2 and an IP address D,
  if D?P1 and D?P2, then D?P1?P2 or D?P2?P1.
• P2 is a subset of P1, D?P2?P1.
• Form Proposition I, there are any two prefix P1
  and P2, they are disjoin or one is the subset of
  another prefix.

11
Definition

• Definition I
• Standalone prefix have no subsets, no supersets.
• Subroot prefix has a subset at least, no
  superset.
• More specific prefix has a superset at least.
• Definition II
• If the prefix in routing tables are represented
  as a Trie data structure, every prefix can has
  the corresponding node in the Trie.
• If there are n prefixes in the path from root to
  the prefix P, we call the prefix P is in Level n.

12
Example
13
The distribution of prefixes
14
Our approach
15
The level-partitioning algorithm
16
Pseudocode for inserting prefix
17
The statistic of routing table
18
The comparison of performance
19
Conclusion

• In the paper, we present a scheme for minimizing
  route update overheads in TCAM forwarding
  engines.
• First, we give the Level-partition algorithm to
  partition routing table and TCAM.
• Second, new update algorithm is proposed to
  reduce the movement overheads of prefix.

20
Conclusion

• On applying these optimizations to a set of real
  route update traces, we find that our scheme can
  produce no more than 100 movements for 10000
  inserting operations, and the worst number of
  movements per update if that of parent prefixes
  in the inserting sequence of prefix.
• Further, when compared to an existing
  optimization algorithm, in the average case, our
  scheme shows a 90 reduction of movement.
• Each Level table has the free spaces. If one of
  them is overflow, it is only one way to
  reconstruct the routing table in TCAM. In future,
  we will find the more efficient scheme to utilize
  memory in TCAM.