简介:FalsesharingisoneofthemostimportantfactorsimpactingtheperformanceofDSM(distributedsharedmemory)systems,Thesingle-Writerapproachissimple,butitcannotavoidtheping-pongeffectofthedatapagethrashing,whilethemultiple-writerapproachiseffectiveforfalsesharingbutwithhighcost.Thispaperproposesanewapproach,calledlimitedmultiple-writer(LMW)tohandlingmultiplewritersinsoftwareDSM.Itdistinguishestwokindsofmultiple-writeraslock-basedformandbarrier-basedform,andhandlesthemwithdifferentpolicies.ItdiscardstheTwinandDiffintraditionalmultiple-writerapproach,andsimplifiestheimplementationofmultiple-writerinsoftwareDSMsystems.TheimplementationofLMWinaCVM(CoherentVirtualMachine)SoftwareDSMsystem,whichisbasedonanetworkofworkstations,isintroduced.EvaluationresultsshowthatforsomeapplicationssuchasSOR(SuccessiveOVer-Relaxation),LU(LowertriangularandUppertriangular),FFT(FastFourierTransformation),andIS(IntegerSorting),LMWprovidesasignificantreductioninexecutiontime(11%,16%,33%and46%)comparedwiththetraditionalmultiple-writerapproachontheplatform.
简介:Forreal-timejammingsignalgenerationindeceivinginversesyntheticapertureradar(ISAR),thetargetcharacteristicsmodulationisalwaysprocessedintheexpensivefieldprogrammablegatearray(FPGA).Duetothelargecomputationalcomplexityofthetraditionalmodulatingoperation,thesizeandstructureofsimulatedfalse-targetarelimited.Withregardtotheprincipleofdechirpinginrangecompressionoflinearfrequencymodulated(LFM)radar,anovelalgorithmnamed'inversedechirping'isproposedfortargetcharacteristicsmodulation.ThisalgorithmonlyneedsonecomplexmultiplierintheFPGAtogeneratethejammingsignalwhentheradarsignalisintercepted,whichcanbeobtainedbymultiplicationofradarsignalsamplingsandtheequivalentdechirpedtargetechointhetimedomain.Asthecomplexsynthesisofdechirpedtargetechocanberealizedbycheapdigitalsignalprocessor(DSP)withintheinterpulsetime,theoverallcostofthejammingequipmentwillbereducedandthefalse-targetsizewillnotbelimitedbythescaleofFPGA.Numericalsimulationsareperformedtoverifythecorrectnessandeffectivenessoftheproposedalgorithm.
简介:用微数组技术的高产量的全球基因表示上的研究曾经产生了大量系统的transcriptome数据。在利用这些异构的数据集的主要挑战是怎么由内部试金的方法使表示侧面正常化。不同非线性、线性的正规化方法被开发了,它实质上依靠在二不同试金之间的真或察觉的对数的褶层变化分布在自然是对称的假设。然而,不对称的基因表达式变化经常被观察,导致非最优的正规化结果并且在后果潜在地到几千假调用。因此,我们明确地调查了不对称的比较transcriptome侧面并且为柔韧、全球的内部试金的正规化用指数的错误工作的加权的否定的秒顺序(NeONORM)开发了正规化。NeONORM高效地抑制规章的事件处理以便在正规化上最小化他们的似是而非的影响的真基因。我们用人工、真的试验性的数据集评估了NeONORM的适用性,哪个证明NeONORM能系统地被用于内部试金、内部条件的比较。