DOI: /jos Tel/Fax: by Journal of Software. All rights reserved.

Transcription

1 ISSN , CODEN RUXUEW Journal of Software, Vol17, No9, September 2006, pp http//wwwjosorgcn DOI /jos Tel/Fax by Journal of Software All rights reserved AM-Trie +,, (, ) AM-Trie A Parallel Multidimensional Packet Classification Algorithm Fitting for Network Processor ZHENG Bo +, LIN Chuang, QU Yang (Department of Computer Science and Technology, Tsinghua University, Beijing , China) + Corresponding author Phn , Fax , bzheng@csnet1cstsinghuaeducn Zheng B, Lin C, Qu Y AM-Trie A parallel multidimensional packet classification algorithm fitting for network processor Journal of Software, 2006,17(9) http//wwwjosorgcn/ /17/1949htm Abstract Nowadays, many high speed Internet applications require high speed multidimensional packet classification algorithms Based on the uniqueness of Network Processor, this paper presents a multidimensional classification algorithm AM-Trie (asymmetrical multi-bit trie) AM-Trie is a high speed, parallel and scalable algorithm and very fit for the multi-thread and multi-core feature of the Network Processor A heuristic field division algorithm is also presented, and it is proved theoretically that it can find out the minimum storage cost solution when the height of the AM-Tire is given Finally, a prototype is implemented based on Intel IXP 2400 Network Processor The performance testing result shows that AM-Trie is a high-speed and scalable algorithm; the throughput of the whole system is influenced little by the size of rules and it can reach 25 Gbps wire speed Key words packet classification; network processor; parallel algorithm; multidimensional; AM-Trie, AM-Trie (asymmetrical multi-bit trie, Trie ),,,, AM-Trie, Intel IXP2400,,, 25Gbps ; ; ;AM-Trie TP393 A Supported by the Major Research Plan of the National Natural Science Foundation of China under Grant No ( ); the National Grand Fundamental Research 973 Program of China under Grant No2003CB ( (973)); the Juniper Research Grant (Juniper ); the Intel IXA University Research Plan (Intel IXA ) Received ; Accepted

2 1950 Journal of Software Vol17, No9, September 2006,,,,Internet (best-effort service) QoS(quality of service), ( ) (content-awareness),, (VPNs) (NAT) (IDS) QoS MPLS IPv6,,,Sigcomm,Infocom [1 5],,,, [6 8],, ( TCAM FPGA),, ( ),,, O(DW), D ;W, IPv6, AM-Trie (asymmetrical multi-bit trie, Trie ), AM-Trie ; AM-Trie,,AM-Trie O(d+h),d,h AM-Trie, O(N 2 ),N ; AM-Trie, AM-Trie ( ), ( ), ;, ;,, Intel IXP2400 ( OC48),, TCP/IP 25Gbps ( 64, 5Mpps) 1 2 AM-Trie 3 AM-Trie 4, 5 1 ASIC, (1) (multi-core) [9],, Cisco PXF,Motorola C-5 DCP ;,, Intel IXP,IBM PowerNP Agere PayloadPlus (2), (multi-thread), 0,Intel IXP2400 8, PCIBM PowerNP

3 AM-Trie 1951 (3) DMA,, DMA,,Intel IXP2400 SRAM (4) [10],,,,,,, (multi-core and multi-thread) AM-Trie, 2 AM-Trie, ( ),,Internet, Internet,,, Trie, IPv4 32 Trie (, 32 ),IPv6 128 Trie,, 2 m Trie Trie, Internet CIDR, 0, Trie, 2 ( 1-bit trie), ( 4, 2-bit, ) AM-Trie 1 k-bit, } 1 67 k 8 } k *,0*,1*,,111*,000,,11 k 1 2 k+1 1, 1,2,,2 k+1 1, (k+1)-bit P*,P p-bit, 2 p +P * 1, 1111* 31, k+1 k, Trie AM-Trie 21 AM-Trie,,, (variable stride) AM-Trie AM-Trie,AM-Trie l, h 1,h 2,,h l,,am-trie 2, h w,w AM-Trie,, ;, ;, ( ), ;,, h i l i=1 i = AM-Trie AM-Trie pnext prule

4 1952 Journal of Software Vol17, No9, September 2006,,, 1 1 AM-Trie, 2-bits Table 1 1 A simple policy table Rule Field R0 (default) * R1 0* R2 01* R3 011* R pnext prule Fig1 AM-Trie creation and search 1 AM-Trie 1 R0, (*), 1, R0 1,R1,R2 2 5 R3, 1 (01) 5, 5 2 (1*),R (*) (01*), R2 (*),R4 R5 R3 011*,, R5 R3, R3 R5, AM-Trie, 2 1 *,,, Trie (asymmetrical multi-bit trie, AM-Trie) AM-Trie,, 0110, 1 01( 5), 5, 2 10( 6), 6,, 1*( 3) 3 R3, 1 22 AM-Trie AM-Trie,,, (push) h i

5 AM-Trie 1953, (, 1* 10 11),,AM-Trie *,,, 2(, ) AM-Trie AM-Trie,, O(h),h AM-Trie AM-Trie, AM-Tire AM-Trie, prule ;,, 0, AM-Trie O(h),h AM-Trie pnext prule Fig2 AM-Trie compression 2 AM-Trie 23 AM-Trie, (ABV) [2] d, AM-Trie,AM-Trie ABV, ( ), AND, N d, AM-Trie d AM-Trie h( h<<w), 1 AM-Trie O(Nlog 2 N) AM-Trie O(Nh), (ABV),, O(Nlog 2 N) 2 AM-Trie O(h+d) AM-Trie ABV h, AM-Trie ; ABV d AM-Trie O(h+d) 3 AM-Trie O(N 2 ) n, P N m-bit, 2 m

6 1954 Journal of Software Vol17, No9, September 2006 ( ), [(h 1) N+1] 2 m sizeof(node)+n sizeof(abv),,sizeof(node) (8, );sizeof(abv) (N-bits) AM-Trie O(N 2 ) 2 AM-Trie,AM-Trie,,,,, Table 2 AM-Trie compared with current algorithms 2 AM-Trie Algorithm Time complexity Space complexity Parallel execution RFC [1] O(logN) O(N d ) No ABV [2] O(W+d) O(N 2 ) Yes HyperCut [3] O(logN) O(N d ) No AM-Trie O(h+d) O(N 2 ) Yes 3 AM-Trie AM-Trie,??, [11],,, 3 BGP [12] ,, 24, Percentage Fig Prefix length (bits) The prefix length distribution of a BGP table 3 BGP 31 AM-Trie AM-Trie ( ), AM-Trie R[N,F 1 (W 1 ),F 2 (W 2 ),,F d (W d )] d,,n ;F i (W i ) i W i 1 F(W), 1-bit trie ( trie,binary trie) i i+j multi-bit trie,, multi-bit trie [nodes(i) leaves(i)] 2 j,,nodes(i) leaves(i) 1-bit trie i i, i+j 2 j ; i+1 i+j i, multi-bit trie [nodes(i) leaves(i)] 2 j

7 AM-Trie R F i (W i ) h W, l 1,l 2,,l h (l 1 +l 2 + +l h =W), h AM-Trie, O(h W 2 ) S(x,y) x 1-bit trie y multi-bit trie 4, W 1-bit trie h multi-bit trie m 1-bit trie h 1 multi-bit trie;, W m h h multi-bit trie Level 0 AM-Trie Level h 1 m Binary trie Cost=S(m,h 1) Level h Fig4 W m For each expanded node Cost=2 W m Recursion analysis of fixed stride AM-Trie 1, S( W, h) = 4 AM-Trie min h 1 m W 1 { S( m, h 1) + [ nodes( m) leaves( m)] 2 W m S(W,1)=2 W (2) S 0~W 0~h, h W, O(h W 2 ) W m, (1), [ nodes( m) leaves( m)] 2, leaves(m) m ; (W m) AM-Trie, AM-Trie, 32 AM-Trie AM-Trie ( ),,, 2 R[N,F 1 (W 1 ),F 2 (W 2 ),,F d (W d )] F i (W i ), h W, l n1,l n2,,l nk (l n1 +l n2 + +l nk =W, 1 n N, 1 k h), h AM-Trie, O(N h W 2 ) C(M,r) M, r AM-Trie height(m) M height(m)=0, nodes(s) nodes(m,s), 1-bit trie s M 5, 1-bit Trie h AM-Trie, G AM-Trie 1, h 1 AM-Trie, m C( M, h) = min 2 + C( M, h 1) 1 m 1 + height ( G) M Nodes( G, m+ 1) C(M,1)=2 1+height(M) (4), AM-Trie } (1) (3)

8 1956 Journal of Software Vo l17, No9, September 2006, O(N W 2 k) AM-Trie Level 0 Binary trie m Cost=2 m Level 1 Level h (at most) T 1 Fig5 4 T 2 T n 1 T n Cost=sum of costs of covering T 1 through T n using h 1 levels Recursion analysis of variable stride AM-Trie 5 Trie Intel IXP2400 AM-Trie,8 1, 4 IP, 3, IXIA 1600 IXP2400, 64 TCP/IP ( IP IP TOS ), 6,6 AM-Trie 3,, MHz DDR SDRAM, AM-Trie 10Mpps 6 32~4096,,, 25Gbps(5Mpps) AM-Trie (,AM-Trie 256K ), ; IXP2400 OC48 (25Gbps),AM-Trie IXP2400, ( ), 7 AM-Trie (, 4096,,, O(N 2 ) ),AM-Trie, RAM (FPGA,TCAM ) IXP M SRAM 2G DRAM, AM-Trie Throughput (Mbps) Number of rules Number of rules Fig6 The throughput of AM-Trie on IXP2400 Fig7 The memory cost of AM-Trie on IXP2400 Memory used (Mbytes) 6 IXP2400 AM-Trie 7 IXP2400 AM-Trie Intel IXP2800,, 10Gbps

9 AM-Trie 1957, 10Gbps( 20M pps) 5 Internet ( TCAM FPGA),, AM-Trie ( Trie ), Trie ; AM-Trie, AM-Trie, ;, AM-Trie O(logW),W, O(N 2 ),N, AM-Trie, AM-Trie ( ) (, Intel IXP2400 ( OC48),,AM-Trie TCP/IP 25Gbps ( 64, 5Mpps), ( IXP2800),, RAM References [1] Gupta P, McKeown N Packet classification on multiple fields In Proc of the ACM SIGCOMM http//tiny- terastanfordedu/~nickm/papers/sigcomm99pdf [2] Baboescu F, Varghese G Scalable packet classification In Proc of the ACM SIGCOMM http//wwwcs ucsdedu/groups/sysnet/miscpapers/p2-baboescupdf [3] Singh S, Baboescu F, Varghese G, Wang J Packet classification using multidimensional cutting In Proc of the ACM SIGCOMM http//wwwsigcommorg/sigcomm2003/papers/p213-singhpdf [4] Baboescu SSF, Varghese G Packet classification for core routers Is there an alternative to cams? In Proc of the IEEE INFOCOM [5] Lakshminarayanan ARK, Venkatachary S Algorithms for advanced packet classification with ternary cams In Proc of the ACM SIGCOMM [6] Network processing forum (npf) http//wwwnpforumorg/ [7] Network systems design conference http//wwwnetworkprocessorscom/ [8] Wolf T, Franklin MS Design tradeoffs for embedded network processors In Proc of the ARCS [9] Shah N Understanding network processors Technical Report, 2001 http//wwwgigascaleorg/pubs/338html [10] McAuley AJ, Francis P Fast routing table lookup using CAMs In INFOCOM (3) [11] Srinivasan V, Varghese G Faster IP lookups using controlled prefix expansion In Measurement and Modeling of Computer Systems [12] BGP routing table analysis reports http//bgppotaroonet/ (1978 ),,,,, (1978 ),,,,, (1948 ),,,,CCF,,, Petri,