基于DVS期限任务包应用功耗感知调度算法研究.pdfVIP

PowerAwareSchedulingofBag-of-TasksApplicationswithDeadlineConstraints

onDVS-enabledClusters

KyongHoonKim,RajkumarBuyyaJongKim

GridComputingandDistributedSystemsLabDept.ofComputerScienceandEngineering

Dept.ofComputerScienceSoftwareEng.PohangUniversityofScienceandTechnology

UniversityofMelbourne,AustraliaSan31,Hyoja-dong,Pohang,Korea

E-mail:{jysh,raj}@.auE-mail:jkim@postech.ac.kr

Abstractdatacenterscanconsumeasmuchelectricityasacity[2,3].

Anotherreasonisrelatedtoreliabilityofsystemsduetoin-

Power-awareschedulingproblemhasbeenarecentissuecreasedtemperaturecausedbylargepowerconsumption.It

inclustersystemsnotonlyforoperationalcostduetoelec-iswellknownthatcomputinginhightemperatureismore

tricitycost,butalsoforsystemreliability.Asrecentcom-error-pronethanoneinappropriateenvironment.Theex-

modityprocessorssupportmultipleoperatingpointsun-derpectedfailurerateofanelectronicdevicedoublesforevery

varioussupplyvoltagelevels,DynamicVoltageScaling10◦CincreasedtemperatureaccordingtotheArrenhius’

(DVS)schedulingalgorithmscanreducepowerconsump-equation[4].Inaddition,theincreasednumberofnodesin

tionbyco