Graphicprocessingunits(GPUs)havebeenwidelyrecognizedascost-efficientco-processorswithacceptablesize,weight,andpowerconsumption.However,adoptingGPUsinreal-timesystemsisstillchallenging,duetothelackinframeworkforreal-timeanalysis.Inordertoguaranteereal-timerequirementswhilemaintainingsystemutilizationinmodernheterogeneoussystems,suchasmulticoremulti-GPUsystems,anovelsuspension-basedk-exclusionreal-timelockingprotocolandtheassociatedsuspension-awareschedulabilityanalysisareproposed.TheproposedprotocolprovidesasynchronizationframeworkthatenablesmultipleGPUstobeefficientlyintegratedinmulticorereal-timesystems.Comparativeevaluationsshowthattheproposedmethodsimproveupontheexistingworkintermsofschedulability.
