摘要
LinuxbasednetworkedPCsclustersarereplacingboththeVMEnonuniformdirectmemoryaccesssystemsandSMPsharedmemorysystemsusedpreviouslyfortheonlineeventfilteringandreconstrucion.ToallowanoptimaluseofthedistributedresourcesofPCclustersanopensoftwareframeworkispresentlybeingdevelopedbasedonadataflowparadigmforeventprocessing.Thisframeworkallowsforthedistributionofthedataofphysicseventsandassociatedcalibrationdatatomultiplecomputersfrommultipleinputsourcesforprocessingandthesubsequentcollectionoftheprocessedeventsatmultipleoutputs.Thebasisofthesystemistheeventrepository,basicallyafirst-infirst-outeventstorewhichmaybereadandwritteninamannersimilartosequentialfileaccess.Eventsarestoredinandtransferredbetweenrepositoriesassuitablylargesequencestoenablehighthroughput.Multiplereaderscanreadsimultaneouslyfromasinglerepositorytoreceiveeventsequencesandmultiplewriterscaninserteventsequencestoarepository,Hencerepositoriesareusedforeventdistributionandcollection.Tosupportsynchronisationoftheeventfolowtherepositoryimplementsbaaiers.Abarriermustbewrittenbyallthewritersofarepositorybeforeanyreadercanreadthebarrier,Areadermustreadabarrierbeforeitmayreceivedatafrombehindit.Onlyafterallreadershavereadthebarrieristhebarrieremovedfromtherepository.Abarriermayalsohaveattacheddata,Inthiswaycalibrationdatacanbedistributedtoallproessuingunits.Therepositoriesareimplementedasmulti-threadedCORBAobjectsinC++andCORMAisusedforalldatatransfers,JobsetupscriptsarewritteninpythonandinteractivestatusandhistogramdisplayisprovidedbyaJavaprogram.JobsrununderthePBSbatchsystemprovidingshareduseofresourcesforonlinetriggering,offlinemassreporcessinganduseranalysisjobs.
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）
