An Encoding Scheme for TCAMBased Packet Classification

1
An Encoding Scheme forTCAM-Based Packet
Classification
Authors Derek Pao, Yiu Keung Li, and Peng
Zhou Publisher ICACT 2006 (The International
Conference on Advanced Communication Technology)
Present Shih-Chin Chang Date Tuesday,
November 24, 2009
Department of Computer Science and Information
Engineering National Cheng Kung University,
Taiwan
2
Outline

• Introduction
• Prefix Inclusion Coding Scheme
• Performance Evaluation
• TCAM Space Requirements
• Incremental Updating

3
Introduction
4
Introduction (cont.)
5
Outline

• Introduction
• Prefix Inclusion Coding Scheme
• Performance Evaluation
• TCAM Space Requirements
• Incremental Updating

6
Prefix Inclusion Coding Scheme

• Let p and q be two distinct address prefixes
  where the length of p is shorter than or equal to
  the length of q.
• Prefixed p and q are either disjoint, or q is
  enclosed by p (i.e. q is a subrange of p).
• Let Cp and Cq be the codewords assigned to p and
  q, respectively. The codeword assignment
  satisfies the following 3 requirements
• A valid codeword must have a non-zero value.
• The inclusion property is preserved. Cq is
  enclosed by Cp iff q is enclosed by p.
• If q is enclosed by p, then Cq must have a
  non-zero suffix extension from Cp.

7
Prefix Inclusion Coding Scheme (cont.)
8
Prefix Inclusion Coding Scheme (cont.)
9
Prefix Inclusion Coding Scheme (cont.)

• Given a sorted list of n prefixes, a tree
  (i-tree) that represents the inclusion prefixes
  can be constructed in O(hn) time where h is the
  depth of the tree.
• Depth of the inclusion tree depends on the
  maximum nesting level prefixes. It has been found
  that the maximum nesting level of prefixes in
  real-life classifiers is about 6. 19
• The minimum codeword length is equal to ?2 of the
  code space occupied by the root. For example, the
  codeword length is 5 bits.
• The overall computation complexity of PIC is
  O(n?n).

19 D. E. Taylor and J. S. Turner, ClassBench
A Packet Classification Benchmark, IEEE INFOCOM
2005,
10
Prefix Inclusion Coding Scheme (cont.)

• In PIC, the address translation is equivalent to
  finding the longest matching prefix in the
  codeword table.
• In the codeword table, a dont care bit in a
  codeword is substituted by a zero.
• For example, the 8-bit input address is 1110 0111
  ? the best matching prefix is J 11100 ? return
  10100
• The codeword lookup can be implement using TCAM.

11
Prefix Inclusion Coding Scheme (cont.)

• In the PIC encoding scheme, if 2 port ranges are
  partially overlapping, one of the 2 ranges will
  be decomposed.
• Using a hybrid encoding method that combines PIC
  with P2C may be avoiding port range decomposition.

12
Prefix Inclusion Coding Scheme (cont.)

• In the hybrid method, partially overlapped ranges
  are grouped into a composite range.
• Within the composite range, B and C are encoded
  using P2C style II, i.e. B is assigned a code
  value of 1x, and C is assigned a code value of
  x1.
• If there is a range R that spans the two basic
  ranges 1-1023 and 1024-65535, then it is not
  recommended to use the hybrid method to encode
  the three ranges. It would be more effective to
  decompose R into two subranges.
• Since the port number only has 16 bits, the
  codeword translation can be implemented using
  direct table lookup with a 64K-entry array.
• In real-life classifiers, the protocol field can
  be represented by a 3-bit or 4-bit codeword.

13
Outline

• Introduction
• Prefix Inclusion Coding Scheme
• Performance Evaluation
• TCAM Space Requirements
• Incremental Updating

14
Performance Evaluation

• The rule set generator developed by Taylor. 19
• There are 3 types of classifiers, namely, the
  access control list (ACL), firewall (FW) and IP
  chain (IPC).

19 D. E. Taylor and J. S. Turner, ClassBench
A Packet Classification Benchmark, IEEE INFOCOM
2005,
15
Performance Evaluation (cont.)
16
TCAM Space Requirements
4
4 / 1
2
3
1
17
Incremental Updating

• Code space management in PIC is conceptually
  similar to the classical blocks packing problem,
  where a given number of variable-sized objects
  are to be packed into fixed-size boxes.
• To facilitate dynamic insertions, up to 4 times
  the required code spaces are allocated to
  internal nodes of the i-tree.
• By doing so, the code length of the field can be
  increased by 2 bits.
• When a new node y is added to the i-tree, the
  minimum code space required by nodes along the
  path from the root to y is recomputed.