摘要
BABAR[1]usestwoformatsforitsdata:ObjectivitydatabaseandROOT[1]files.Thisposterconcernsthedistributionofthelatter-forObjectivitydatasee[3].TheBABARanalysisdataisstoredinROOTfiles-oneperphysicsrunandanalysisselectionchannel-maintainedinalargedirectorytree,CurrentlyBABARhasmorethan4.5TBytesin200,000ROOTfiles.Thisdatais(mostly)producedatSLAC,butisrequiredforanalysisatuniversitiesandresearchcentresthroughouttheUSandEurope.TwobasicproblemsconfrontuswhenweseektoimportbulkdatafromSLACtoaninstitute'slocalstorageviathenetwork.Wemustdeterminewhichfilesmustbeimported(dependingonthelocalsiterequirementsandwhichfileshavealreadybeenimported),andwemustmaketheoptimumuseofthenetworkwhentransferringthedata,Basicftp-liketools(ftp,scp,etc)donotattempttosolvethefirstproblem.Moresophisticatedtoolslikersync[4],thewidely-usedmirror/synchronisationprogram,comparelocalandremotefilesystems,checkingforchanges(basedonfiledate,sizeand,ifdesired,anelaboratechecksum)inordertoonlycopynewormodifiedfiles,Howeverrsyncallowsforonlylimitedfileselection.Alsowhen,asinBABAR,anextremelylargedirectorystructuremustbescanned,rsynccantakeseveralhoursjusttodeterminewhichfilesneedtobecopied.Althoughrsync(andscp)provideson-the=flycompression,itdoesnotallowustooptimisethenetworktransferbyusingmultiplestreams,abjustingtheTCPwindowsizeorseparatingencryptedauthenticationfromunencrypteddatachannels.
出版日期
2001年01月11日（中国期刊网平台首次上网日期，不代表论文的发表时间）
