《面向深度包检测的DFA细粒度并行匹配方法-论文》.pdfVIP

计算机研究与发展 DOI：10．7544／issn1000—1239．2014．2O120741 51(5)：1061—1070，2014 JournalofComputerResearchandDevelopmen t — — 面 向深度包检测的DFA细粒度并行匹配方法 刘兴奎 。邵宗有 “刘新春 孙凝晖 (中国科学院计算技术研究所高性能计算机研究中心 北京 100190) 。(中国科学院大学 北京 100049) (北京科技大学信息工程学院 北京 100083) (曙光信息产业 (北京)有限公司 北京 100094) (xingkuiliu@ncic．ac．cn) Fine．GrainedParallelRegularExpressionM atchingforDeep PacketInspection IiuXingkui一，ShaoZongyou。一，LiuXinchun ，andSunNinghui (Hig Per_厂。r ncPCom rResearchCenPr，Institut。_厂ComputingTech竹。1。gY，ChineseAcadem3，。厂sc绷cP，B jing 100190) 。((_ Pr ofChineseAcademyofSciences，Beijing100O49) 。(Sco。l。fInf。rma￡。”Engineering，University。fSci cendTechnoZ。gYBeiji”g，Beijig100083) (SugonCompany，Beijing100094) Abstract RegularexpressionmatchingplaysanimportantroleinmanycriticalnetworkaPPllcatons· Deterministicfiniteautomata (DFA)isan effectiveway toimplementregularexpresmo“matchmg， however，DFAs’inherentsequentialstatetransitionmakesthem impracticalforhigh—speedbackbon networks．In thispaper，anovelfine—grained parallelDFA，calledLBDFA (Loopback DFA)，is pr0posedtoimprovethematchingperformanceofDFAs．Themethodisbasedontheobservationthat mosttransitions occur among a smallnumber of states while other states are rarelY accessed． Furthermore，thefrequently traversed states，calledLoopback statesin thispaper，USuallyremam unchangedf0r a largenumberofconsecutive inputcharacters in theprocessoistatetranstons· Theref0rearemarkableimprovementcanbeachievedbyparallelizingtheconsecutivestatetrans。tons onLooDback states．A Bloom filterisemployedtoeliminatethetemporarydeviation ntranstons n ordertofurtherimprovetheparallelism ofLBDFAs．Experimentalresultsonrulesetsfrom L ／一±lter andSnorfshow thattheLBDFA canmeetthedemandofregularexpr