{"title":"用于多处理器运行时适应的基础结构","authors":"J. Appavoo, Kevin Hui, M. Stumm, R. Wisniewski, D. D. Silva, O. Krieger, Craig A. N. Soules","doi":"10.1145/582128.582130","DOIUrl":null,"url":null,"abstract":"Runtimeadaptationanddynamicrecon gurationallowasystemtodynamicallyswapinthemostappropriateimple-mentationofitscomp onentsbasedoncurrentorexp ecteduse,andtoselectivelyupgradecomp onentswithbug,secu-rity,orp erformance xeswithoutdown-time.Hot-swappingistheactofreplacinganactivesystemcomp onent'sim-plementationwithanewordi erentimplemenwhilemaintainingavailabilityofthecomp onent'sfunctionalit.Thispap erdescrib esamechanism tohot-swapsoftwarecom-p onentswithintheK421op eratingsystem.Wehavusedthiscapabilitytoimprovep erformanceunderaarietofconditions.K42isaresearchop eratingsystemforcache-coherentsharedmemorymultipro cessorsdesignedtoachievego o dscalabilityonawiderangeofworkloads.InK42,eachvir-tualandphysicalresource,e.g.,op en le,memoryregion,pagetable,ismanagedbyseparate, ne-granularit,ob jectinstances.Eachob jectinstancemayb ecustomized(di er-entsub-class).Thismo delprovidesthestandardsoftwareengineeringb ene ts,butmoreimp ortantly:1)allowsp er-formancecustomizationonanob ject-by-ob jectbasisand2)allows,onamultipro cessor,indep endentaccessestob edirectedtoindep endentinstancesandpro ceedinparallelthuseliminatingsharedmemoryaccessesandsynchroniza-tion,whicharefundamentalbarrierstomultipro cessorp er-formance.Anop eratingsystemisaparticularlydemandingenvi-ronmentasitneedstosatisfymultipleconcurrenclientswhoseresourcedemandsmayb eato dds.Forexample,multipleclientsmaysimultaneouslyaccessa lewithdif-ferentusagepatterns.Supp ortingmultipro cessorspresentsadditionalchallengesforop eratingsystems,andoftenimple-mentationsthatarerequiredforscalablep erformancehaveworseunipro cessorb ehavior.Toprovideb othunipro cessorandmultipro cessorcomp o-nentswithinasingleob jectmo del,K42'sob jectsareim-1http://www.research.ibm.com/K42","PeriodicalId":326554,"journal":{"name":"Workshop on Self-Healing Systems","volume":"13 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2002-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"37","resultStr":"{\"title\":\"An infrastructure for multiprocessor run-time adaptation\",\"authors\":\"J. Appavoo, Kevin Hui, M. Stumm, R. Wisniewski, D. D. Silva, O. Krieger, Craig A. N. Soules\",\"doi\":\"10.1145/582128.582130\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Runtimeadaptationanddynamicrecon gurationallowasystemtodynamicallyswapinthemostappropriateimple-mentationofitscomp onentsbasedoncurrentorexp ecteduse,andtoselectivelyupgradecomp onentswithbug,secu-rity,orp erformance xeswithoutdown-time.Hot-swappingistheactofreplacinganactivesystemcomp onent'sim-plementationwithanewordi erentimplemenwhilemaintainingavailabilityofthecomp onent'sfunctionalit.Thispap erdescrib esamechanism tohot-swapsoftwarecom-p onentswithintheK421op eratingsystem.Wehavusedthiscapabilitytoimprovep erformanceunderaarietofconditions.K42isaresearchop eratingsystemforcache-coherentsharedmemorymultipro cessorsdesignedtoachievego o dscalabilityonawiderangeofworkloads.InK42,eachvir-tualandphysicalresource,e.g.,op en le,memoryregion,pagetable,ismanagedbyseparate, ne-granularit,ob jectinstances.Eachob jectinstancemayb ecustomized(di er-entsub-class).Thismo delprovidesthestandardsoftwareengineeringb ene ts,butmoreimp ortantly:1)allowsp er-formancecustomizationonanob ject-by-ob jectbasisand2)allows,onamultipro cessor,indep endentaccessestob edirectedtoindep endentinstancesandpro ceedinparallelthuseliminatingsharedmemoryaccessesandsynchroniza-tion,whicharefundamentalbarrierstomultipro cessorp er-formance.Anop eratingsystemisaparticularlydemandingenvi-ronmentasitneedstosatisfymultipleconcurrenclientswhoseresourcedemandsmayb eato dds.Forexample,multipleclientsmaysimultaneouslyaccessa lewithdif-ferentusagepatterns.Supp ortingmultipro cessorspresentsadditionalchallengesforop eratingsystems,andoftenimple-mentationsthatarerequiredforscalablep erformancehaveworseunipro cessorb ehavior.Toprovideb othunipro cessorandmultipro cessorcomp o-nentswithinasingleob jectmo del,K42'sob jectsareim-1http://www.research.ibm.com/K42\",\"PeriodicalId\":326554,\"journal\":{\"name\":\"Workshop on Self-Healing Systems\",\"volume\":\"13 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2002-11-18\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"37\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Workshop on Self-Healing Systems\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1145/582128.582130\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Workshop on Self-Healing Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/582128.582130","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 37
An infrastructure for multiprocessor run-time adaptation
Runtimeadaptationanddynamicrecon gurationallowasystemtodynamicallyswapinthemostappropriateimple-mentationofitscomp onentsbasedoncurrentorexp ecteduse,andtoselectivelyupgradecomp onentswithbug,secu-rity,orp erformance xeswithoutdown-time.Hot-swappingistheactofreplacinganactivesystemcomp onent'sim-plementationwithanewordi erentimplemenwhilemaintainingavailabilityofthecomp onent'sfunctionalit.Thispap erdescrib esamechanism tohot-swapsoftwarecom-p onentswithintheK421op eratingsystem.Wehavusedthiscapabilitytoimprovep erformanceunderaarietofconditions.K42isaresearchop eratingsystemforcache-coherentsharedmemorymultipro cessorsdesignedtoachievego o dscalabilityonawiderangeofworkloads.InK42,eachvir-tualandphysicalresource,e.g.,op en le,memoryregion,pagetable,ismanagedbyseparate, ne-granularit,ob jectinstances.Eachob jectinstancemayb ecustomized(di er-entsub-class).Thismo delprovidesthestandardsoftwareengineeringb ene ts,butmoreimp ortantly:1)allowsp er-formancecustomizationonanob ject-by-ob jectbasisand2)allows,onamultipro cessor,indep endentaccessestob edirectedtoindep endentinstancesandpro ceedinparallelthuseliminatingsharedmemoryaccessesandsynchroniza-tion,whicharefundamentalbarrierstomultipro cessorp er-formance.Anop eratingsystemisaparticularlydemandingenvi-ronmentasitneedstosatisfymultipleconcurrenclientswhoseresourcedemandsmayb eato dds.Forexample,multipleclientsmaysimultaneouslyaccessa lewithdif-ferentusagepatterns.Supp ortingmultipro cessorspresentsadditionalchallengesforop eratingsystems,andoftenimple-mentationsthatarerequiredforscalablep erformancehaveworseunipro cessorb ehavior.Toprovideb othunipro cessorandmultipro cessorcomp o-nentswithinasingleob jectmo del,K42'sob jectsareim-1http://www.research.ibm.com/K42