File system designers continue to look to new architectures to improve scalability. Object-based storage diverges from server-based (e. g., NFS) and SAN-based storage systems by coupling processors and memory with disk drives, delegating low-level allocati
Ceph:AScalable,High-PerformanceDistributedFileSystem
SageA.Weil
ScottA.BrandtEthanL.MillerDarrellD.E.Long
CarlosMaltzahn
UniversityofCalifornia,SantaCruz
{sage,scott,elm,darrell,carlosm}@cs.ucsc.edu
Abstract
Filesystemdesignerscontinuetolooktonewarchitec-turestoimprovescalability.Object-basedstoragedi-vergesfromserver-based(e.g.,NFS)andSAN-basedstoragesystemsbycouplingprocessorsandmemorywithdiskdrives,delegatinglow-levelallocationtoobjectstoragedevices(OSDs)anddecouplingI/O(read/write)frommetadata( leopen/close)operations.Evenre-centobject-basedsystemsinheritdecades-oldarchitec-turalchoicesgoingbacktoearlyUNIX lesystems,how-ever,limitingtheirabilitytoeffectivelyscaletohundredsofpetabytes.
WehavedevelopedCeph,adistributed lesystemthatprovidesexcellentperformanceandreliabilitywhilepromisingunprecedentedscalability.Cephmaximizestheseparationbetweendataandmetadatamanagementbyreplacingallocationtableswithapseudo-randomdatadistributionfunction(CRUSH)designedforheteroge-neousanddynamicclustersofunreliableOSDs.WeleverageOSDintelligencetodistributedatareplication,failuredetectionandrecoverywithsemi-autonomousOSDsrunningaspecializedlocalobjectstorage lesys-tem(EBOFS).Cephisbuiltaroundadynamicdistributedmetadataclusterthatprovidesextremelyef cientmeta-datamanagementandseamlesslyadaptstoawiderangeofgeneralpurposeandscienti ccomputing lesystemworkloads.PerformancemeasurementsunderavarietyofworkloadsshowthatCephhasexcellentI/Operfor-manceandscalablemetadatamanagement,supportingmorethan250,000metadataoperationspersecond.
1Introduction
Systemdesignershavelongsoughttoimprovetheper-formanceof lesystems,whichhaveprovedcriticaltotheoverallperformanceofanexceedinglybroadclassofapplications.Thescienti candhigh-performancecom-putingcommunitiesinparticularhavedrivenadvancesintheperformanceandscalabilityofdistributedstor-1
agesystems,typicallypredictingmoregeneralpurposeneedsbyafewyears.Traditionalsolutions,exempli edbyNFS[18],provideastraightforwardmodelinwhichaserverexportsa lesystemhierarchythatclientscanmapintotheirlocalnamespace.Althoughwidelyused,thecentralizationinherentintheclient/servermodelhasprovenasigni cantobstacletoscalableperformance.Morerecentdistributed lesystemshaveadoptedar-chitecturesbasedonobject-basedstorage,inwhichcon-ventionalharddisksarereplacedwithintelligentobjectstoragedevices(OSDs)whichcombineaCPU,networkinterface,andlocalcachewithanunderlyingdiskorRAID[3,28,7,6,31].OSDsreplacethetraditionalblock-levelinterfacewithoneinwhichclientscanreadorwritebyterangestomuchlarger(andoftenvariablysized)namedobjects,distributinglow-levelblockallo-cationdecisionstothedevicesthemselves.Clientstyp-icallyinteractwithametadataserver(MDS)toperformmetadataoperations(open,rename,etc.),whilecommu-nicatingdirectlywithOSDstoperform leI/O(readsandwrites),signi cantlyimprovingoverallscalability.Systemsadoptingthismodelcontinuetosufferfromscalabilitylimitationsduetolimitedornodistributionofthemetadataworkload.Continuedrelianceontradi-tional lesystemprincipleslikeallocationlistsandin-odetablesandareluctancetodelegateintelligencetotheOSDshavefurtherlimitedscalabilityandperformance,andincreasedthecostofreliability.
WepresentCeph,adistributed lesystemthatpro-videsexcellentperformanceandreliabilitywhilepromis-ingunparalleledscalability.Ourarchitectureisbasedontheassumptionthatsystemsatthepetabytescalearein-herentlydynamic:largesystemsareinevitablybuiltin-crementally,nodefailuresarethenormratherthantheexception,andthequalityandcharacterofworkloadsareconstantlyshiftingovertime.
Cephdecouplesmetadataanddataoperationsbyelim-inating leallocationtablesandreplacingthemwithgen-eratingfunctions.ThisallowsCephtoleveragethein-
Ceph file system_IT/计算机_专业资料。distributed file system ,分布式文件系统 linuxCeph:?Reliable,?Scalable,? and?High?performance? Distributed?Storage Sage...
E. Long , Carlos Maltzahn, Ceph: a scalable, high-performance distributed file system, Proceedings of the 7th conference on USENIX Symposium on Operating ...
Ceph :A Scalable, High-... 14页 免费 Can ...High-Performance Distributed File System Roman ... Switzerland u cca@zurich.ibm.com Abstract ...
Abstract Ceph A Scalable... 14页 免费 ABSTRACT...Distributed File System John Bent, Douglas Thain,...The prithe performance bene?ts of explicit control...
SCALABLE, AND HIGH-PERFORMANCE DISTRIBUTED STORAGE.... . . 189 xiv Abstract Ceph: Reliable, ...2.1 Local File Systems File system design over ...
Luster file system Ceph: A Scalable, High-Performance Distributed File System Google File System 系统架构 ? ? ? ? ? C/S体系结构 共享存储SAN架构 基于...
Ceph Distributed Storage System提供企业级支持 ceph...Controlled Replication Under Scalable Hashing(CRUSH)... device across the cluster for high performance. ...
(Extent and B-tree-based Object File System [...叫做 Reliable Autonomic Distributed Object Storage(...Controlled Replication Under Scalable Hashing(CRUSH)...
Abstract Ceph A Scalable... 暂无评价 15页 免费 Abstract Distributed Par....t of mechanisms for improving ?le system performance currently available ...
E. Long, and Carlos Maltzahn, ``Ceph: A Scalable, High-Performance, ...Lustre is a distributed object-based file system, as is the Panasas file ...
热门文档