Storage device performance prediction with CART models

Storage device performance prediction is a key element of self-managed storage systems and application planning tasks, such as data assignment. This work explores the application of a machine learning tool, CART models, to storage device modeling. Our appr

StorageDevicePerformancePredictionwithCART

Models

MengzhiWang,KinmanAu,AnastassiaAilamaki,

AnthonyBrockwell,ChristosFaloutsos,andGregoryR.Ganger

CMU-PDL-04-103

March2004

ParallelDataLaboratoryCarnegieMellonUniversityPittsburgh,PA15213-3890

Abstract

Storagedeviceperformancepredictionisakeyelementofself-managedstoragesystemsandapplicationplanningtasks,suchasdataassignment.Thisworkexplorestheapplicationofamachinelearningtool,CARTmodels,tostoragedevicemodeling.Ourapproachpredictsadevice’sperformanceasafunctionofinputworkloads,requiringnoknowledgeofthedeviceinternals.WeproposetwousesofCARTmodels:onethatpredictsper-requestresponsetimes(andthenderivesaggregatevalues)andonethatpredictsaggregatevaluesdirectlyfromworkloadcharacteristics.Afterbeingtrainedonourexperimentalplatforms,bothprovideaccurateblack-boxmodelsacrossarangeoftesttracesfromrealenvironments.Experimentsshowthatthesemodelspredicttheaverageand90thpercentileresponsetimewithanrelativeerroraslowas16%,whenthetrainingworkloadsaresimilartothetestingworkloads,andinterpolatewellacrossdifferentworkloads.

Acknowledgements:WethankthemembersandcompaniesofthePDLConsortium(includingEMC,Hewlett-Packard,Hitachi,HitachiGlobalStorageTechnologies,IBM,Intel,LSILogic,Microsoft,NetworkAppliance,Oracle,Panasas,Seagate,Sun,andVeritas)fortheirinterest,insights,feedback,andsupport.WethankIBMforpartlyfundingthisworkthroughaCASstudentfellowshipandafacultypartnershipaward.ThisworkisfundedinpartbyNSFgrantsCCR-0205544,IIS-0133686,BES-0329549,IIS-0083148,IIS-0113089,IIS-0209107,andIIS-0205224.WewouldalsoliketothankEnoThereska,MikeMesnier,andJohnStrunkfortheirparticipationanddiscussionintheearlystageofthisproject.