欢迎来到天天文库
浏览记录
ID:20421619
大小:63.50 KB
页数:8页
时间:2018-10-13
《基于authorware的交互课件的设计与制作》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、TheHadoopDistributedFilesystemWhenadatasetoutgrowsthestoragecapacityofasinglephysicalmachine,itbecomesnecessarytopartitionitacrossanumberofseparatemachines.Filesystemsthatmanagethestorageacrossanetworkofmachinesarecalleddistributedfilesystems.Sincetheyarenetwork-based,allthecomplicationsofne
2、tworkprogrammingkickin,thusmakingdistributedfilesystemsmorecomplexthanregulardiskfilesystems.Forexample,oneofthebiggestchallengesismakingthefilesystemtoleratenodefailurewithoutsufferingdataloss.HadoopcomeswithadistributedfilesystemcalledHDFS,whichstandsforHadoopDistributedFilesystem.(Youmays
3、ometimesseereferencesto“DFS”—informallyorinolderdocumentationorconfiguration—whichisthesamething.)HDFSisHadoop’sflagshipfilesystemandisthefocusofthischapter,butHadoopactuallyhasageneralpurposefilesystemabstraction,sowe’llseealongthewayhowHadoopintegrateswithotherstoragesystems(suchasthelocal
4、filesystemandAmazonS3).TheDesignofHDFSHDFSisafilesystemdesignedforstoringverylargefileswithstreamingdataaccesspatterns,runningonclustersoncommodityhardware.Let’sexaminethisstatementinmoredetail:Verylargefiles“Verylarge”inthiscontextmeansfilesthatarehundredsofmegabytes,gigabytes,orterabytesin
5、size.ThereareHadoopclustersrunningtodaythatstorepetabytesofdata.*StreamingdataaccessHDFSisbuiltaroundtheideathatthemostefficientdataprocessingpatternisawrite-once,read-many-timespattern.Adatasetistypicallygeneratedorcopiedfromsource,thenvariousanalysesareperformedonthatdatasetovertime.Eachan
6、alysiswillinvolvealargeproportion,ifnotall,ofthedataset,sothetimetoreadthewholedatasetismoreimportantthanthelatencyinreadingthefirstrecord.CommodityhardwareHadoopdoesn’trequireexpensive,highlyreliablehardwaretorunon.It’sdesignedtorunonclustersofcommodityhardware(commonlyavailablehardwareavai
7、lablefrommultiplevendors†)forwhichthechanceofnodefailureacrosstheclusterishigh,atleastforlargeclusters.HDFSisdesignedtocarryonworkingwithoutanoticeableinterruptiontotheuserinthefaceofsuchfailure.ItisalsoworthexaminingtheapplicationsforwhichusingHDF
此文档下载收益归作者所有