1. 1 An innovative low-cost Classification Scheme for combined multi-Gigabit IP and Ethernet Networks Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C. Authors ： Ioannis Papaefstathiou, and Vassilis Papaefstathiou. Publisher ： IEEE ICC 2006 Present ： Yu-Tso Chen Date ： October, 15, 2008

2. 2 Outline 1. Introduction 2. DLCE 3. 4. Performance Analysis

3. 3 Introduction Double Layer Classification Engine (DLCE) Uses a new hashing scheme and internal replacement of MAC Vendor IDs at the Ethernet layer Same hashing scheme together with an innovative trie-based engine for the IP classificaton

4. 4 Double Layer Classification Engine Every rule in the ruleset is associated to a number called FlowID 15-bit FlowIDs, translating to 32K unique and independent network flows

5. 5 Original MAC Address Hashing Hval = MAC[31:0] MAC_TBL index MAC_TBL – Hold the FlowID of the associated rule.

6. 6 MAC_TBL entries format and memory organization

7. 7

8. 8 MAC Vendor Replacement 24-bit vendor address space of the MAC addresses Fewer than 8000 vendors are active instead of the 2 24 possible Each MAC-address rule handled by our system is 37-bits(13+24bit) long address space can be handled by a 24-to-13 decoder

9. 9 Vendor Replacement MAC Address Hashing 21 Bit

10. 10 IP Layer Classification Uses strides and multi-bit trie nodes in order to traverse several bits at a time and produce fast lookups. Its nodes are represented with bitmaps that can be processed fast in hardware and require small storage.

11. 11 Bitmaps PrefixNO = PrefixValue + 2 PrefixLength – 1 TrieBMP – Efficiently the information about the subtries PrefixBMP – The information about the active prefixes belonging to subtrie

12. 12 SubtrieNO = PrefixValue[0:3] + 1

13. 13 Bit position of the actual prefix in a specific subtrie 2 PrefixLength – 1 PrefixValue

14. 14 Trie data structure example

15. 15 Inside the two subtries the candidate prefixes are four Priority High

16. 16 DLCE Table The interval 0-7 we have a single entry for root called TBL0 Interval 8-15 we have 2 8 possible roots (TBL8) Interval 16-23 we use a 2 16 =65536 table called TBL16

17. 17 DLCE Table (cont.) Interval 24-31 use 2 16 entries in table TBL24 (indexing by hashing the first 24-bits) 32 bit prefixes we use only 2 12 entries in table TBL32

18. 18 Sequence of lookups TBL16 → TBL24 → TBL32 → TBL8 → TBL0 If after TBL32 a match was produced then our lookup process does not proceed to the next tables.

19. 19 Performance

20. 20

21. 21

22. 22 MAC Address Hashing MAC_TBL – Hold the FlowID of the associated rule. EXACT_TBL – The number of entries in each variable size block. (Defined by the number of rules that collide within a specific entry)