Agenda
• PART1
-ParallelComputingToolbox
-TaskParallelization
-DataParallelization
-Batchmode
-Interactivewithpmode
• PART2
-ConfigureMATLABandPCTonPC
-Batchscript
-RunPCTonremoteclusters(hound,orca)
-MoreexamplesandDemos
Parallel Computing Toolbox
• Goal?
ParallelComputingToolboxletsyousolvecomputationallyanddata-
intensiveproblemsusingMATLABonmulti-coreandmultiprocessor
computers.
• Thewayitdoes:
Parallelprocessingconstructssuchasparallelfor-loopsandcodeblocks,
distributedarrays,parallelnumericalalgorithms,andmessage-passing
functionsletyouimplementtask-anddata-parallelalgorithmsinMATLABat
ahighlevelwithoutprogrammingforspecifichardwareandnetwork
architectures.
• Results
ConvertingserialMATLABapplicationstoparallelMATLABapplications
requiresfewcodemodificationsandnoprogramminginalow-levellanguage.
Youcanrunyourapplicationsinteractivelyonalocalmachine,orinbatch
environmentsonaremoteserver.
ParallelComputingToolboxKeyFeatures
• Supportfordata-parallelandtask-parallelapplicationdevelopment
• Abilitytoannotatecodesegmentswithparfor(parallelfor-loops)and
spmd(singleprogrammultipledata)forimplementingtask-anddata
-parallelalgorithms
• High-levelconstructssuchasdistributedarrays,parallelalgorithms,
andmessage-passingfunctionsforprocessinglargedatasetson
multipleprocessors
• Abilitytorun8workerslocallyonamulti-coredesktop(R2010b)
(defaulttothenumberofcoresavailableonaPC)
• IntegrationwithMATLABDistributedComputingServerforcluster-
basedapplicationsthatuseanyscheduleroranynumberofworkers
• Interactiveandbatchexecutionmodes
PCTArchitecture(client-server)