网络处理器设计的若干关键技术研究

《网络处理器设计的若干关键技术研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、国防科学技术大学博士学位论文网络处理器设计的若干关键技术研究姓名：张晓明申请学位级别：博士专业：计算机科学与技术指导教师：张民选;孙志刚20060401国防科学技术大学研究生院博士学位论文ABSTRACTWitllthedevelopmentofnetworkapplications．networkdevicesneedmoreintelligentprocessingcapability．Thisrequiresnetworkdevicestohavevariousfunctions(e．g．multi-layerswitching，

2、Securityprocessingandtrafficmanagement)aswellaspowerfulprotocolprocessingcapabilityandprogrammability,SOthatthenovelnetworkservicesCanbequicklydeployedandconfiguredinthesenetworkdevices．Thusnetworkprocessors(NPs)whicharebasedonthetechnologyofApplicationSpecificInstructio

3、nProcessor(ASIP)emergetimelyandarewidelyusedinnetworkdomainstomeettheserequirements．NPshavealreadybecomeoneofthecoredevicesinthenext-generationInternet．ThisdissertationfocusesontheissuesofsystemdesignandimplementationofNPs．TheearlydesignmethodandperformanceevaluationofNP

4、sarepresentedonthestandpointofsystemdesign，andseveralkeyimplementationtechnologiesofNPsareinvestigatedin·depthinthisdissertation．Themaincontributionsofthedissertationareasfollows：(1)AimingattheoptimaldecisionandperformanceevaluationofsystemdesigninNPs，theYinHeNetworkProc

5、essorDesignFramework(YH-NPDF)isconstructedaccordingtothecharacterizationofMulti-ProcessorSystemonChip(MPSoC)designandrequirementsofnetworkpacketprocessing．TheYH-NPDFisbasedontheideaofplatform-baseddesign．ItadoptstheReactiveDataflowProcessNetwork(RDPN)modeltodescribenetwo

6、rkapplicationsandestablishestheparameterizedmodelofNPs’hardwareresources，whereapplicationmodelismappedintotheparameterizedarchitecturemodelofNPstoevaluatetheNPperformance．Theglobalannealinggeneticalgorithmisusedtoacceleratethesearchofdesignspaceandtooptimizethedesigndeci

7、sionofNPsystem．TheYH—NPDFCanbeusedtomodelnetworkapplicationsandhardware。resourcesandsupportoptimaldecisiontomeettherequirementsforintelligentpacketprocessinginearlysystemdesignofNPs．(2)Innetworkprocessorsbasedonparallelprocessingelements(PEs)，apacketparallelschedulingalg

8、orithmbasedonFuzzyFeedbackControlLoop(F2CL)isproposed．ThisalgorithmusesF2CLschemestoimprovethedegreeofl