- 1、本文档共23页,可阅读全部内容。
- 2、原创力文档(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
- 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
计算机研究与发展DOI:10.7544/issn1000-1239.202440190
JournalofComputerResearchandDevelopment61(8):1945?1956,2024
Graph4Cache:一种用于缓存预取的图神经网络模型
尚晶武智晖肖智文张逸飞
(中国移动信息技术中心北京100033)
(中移动信息技术有限公司北京100033)(shangjing@)
Graph4Cache:AGraphNeuralNetworkModelforCachePrefetching
ShangJing,WuZhihui,XiaoZhiwen,andZhangYifei(ChinaMobileInformationTechnologyCenter,Beijing100033)(ChinaMobileInformationTechnologyCo.,Ltd.,Beijing100033)
AbstractMostcomputingsystemsutilizecachingtoreducedataaccesslatency,speedupdataprocessingandbalanceserviceload.Thekeytocachemanagementistodeterminetheappropriatedatatobeloadedintoordiscardedfromthecache,aswellastheappropriatetimingforcachereplacement,whichiscriticaltoimprovingcachehitrate.Theexistingcachingschemesfacewithtwoproblems:Inreal-timeandonlinecachingscenarios,itisdifficulttodiscerntheheatinformationofuseraccesstodatawhileignoringthecomplexhigh-orderinformationamongdata-access-sequences.Inthispaper,weproposeaGNN-basedcacheprefetchingnetworknamedGraph4Cache.Wemodelasingleaccesssequenceintoadirectedgraph(ASGraph),wherevirtualnodesareusedtoaggregatethefeaturesofallnodesingraphandrepresentthewholesequence.Thenacrosssequenceundirectedgraph(CSGraph)isconstructedfromthevirtualnodesofASGraphstolearncross-sequencefeatures,whichgreatlycomplementsthelimiteditemtransitionsinasinglesequence.Byfusingtheinformationofthesetwographs,welearnthehigh-ordercorrelationsamongsequencesandgetabundantuserintents.Experimentalresultsonmultiplepublicdatasetsdemonstratetheeffectivenessofthismethod.Graph4Cacheoutp
文档评论(0)