1. 1 Recursively Partitioned Static IP Router Table Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C. Authors ： Wencheng Lu, Sartaj Sahni Publisher ： ISCC 2007 Present ： Kuang-Ying Ho 何冠穎 Date ： 2007/11/06(Tue.)

2. 2 Introduction A method–recursive partitioning–to partition a static IP router table so that when each partition is represented using a base structure such as a multibit trie (MST) or a hybrid shape shifting trie (HSST). Reduce both Total memory required for router table. Number of memory access. Compare with popular front-end table methed.

3. 3 *Shape Shifting Trie K: node size in STT SBMShape Bitmap2K bit IBMInternal Bitmap (valid bit) K bits EBMExternal Bitmap (exit point) K+1 bits

4. 4 Recursive Partitioning R s : 3 R First-level partitions of T partition L(R)the auxiliary partition s : stride, 1 ≤ s ≤ T.height+1 s : 2

5. 5 Q(N) : bit string Index of ST(N) Data structure

6. 6 Hash table - Entry types R s : 2

7. For first level s : stride ht : address of first hash table entry h : perfect hash function d : destination IP 7 Lookup qu

8. Incorporating Leaf Pushing 8 R

9. 9 Lookup after leaf pushing

10. 10 Dynamic Programming Recurrence B(N, l, r) be the minimum memory required to represent levels 0 through l of the subtree of T rooted at N by a base structure such as MBT or HSST take no more than r memory accesses. H(N, l) be the memory required for a stride l hash table for the paths from node N of T to nodes in D l (N) C(N, l, r) be the minimum memory required by a recursively partitioned representation of the subtrie defined by levels 0 through l of ST(N). r = 4, 5 0 < l ≤ N.height N N Q N.height l Recurrences for B may be obtained from Sahni and Kim [12] for fixed- and variable-stride MBTs and Lu and Sahni [6] for HSSTs.

11. Optimization 11 A. when auxiliary partitions L(R) are restricted to be resented by base structures, the memory requirement is reduced. B. either a hash table or a simple array with 2 l entries can be use when the partition stride is l.

12. 12 Implementation For benchmarking purposes we assumed that the router table will reside on a QDRII SRAM (dual burst), which supports the retrieval of 72 bits of data with a single memory access. We considered two hash-table designs–36 bit and 72 bit.

13. 13 Implementation for IPv4 In the 36-bit design for IPv4, we allocated 36 bits to each hash entry with: 8 bits for Q(N), 2 bits for the stride of the next-level partition (5-8), 8 bits for the mask, 17 bits for the pointer. In the 72-bit design for IPv4, we allocated 72 bits for each hash-table entry with 17 bits for Q(N), 5 bits for the stride of the next-level partition (1-17), 17 bits for the mask, 19 bits for the pointer

14. 14 Implementation for IPv6 In the 36-bit design for IPv6, we allocated 36 bits to each hash entry with: 7 bits for Q(N), 2 bits for the stride of the next-level partition (4-7), 7 bits for the mask, 19 bits for the pointer. In the 72-bit design for IPv6, we allocated 72 bits for each hash-table entry with 17 bits for Q(N), 5 bits for the stride of the next-level partition (1-17), 17 bits for the mask, 19 bits for the pointer

15. 15 Performance for IPv4

16. 16 Performance for IPv6

17. 17 Contributions