Partially Non-Preemptive Dual Priority Multiprocessor Scheduling

Abstract

AbstractWe propose a partially non-preemptive dual priority scheduling algorithm (PNPDP) for multiprocessors. In dual priority scheduling, each task has two fixed priorities. When a job is released, it executes at its task’s lower priority. After some fixed amount of time, its priority is promoted. Our approach is to prevent lower priority jobs from preempting one another. We use the tasks’ Worst Case Response Times to determine when a promotion must occur in order to guarantee all deadlines will be met. During execution, this promotion time is adjusted to extend non-preemptive execution of lower priority tasks whenever possible. Tasks executing at their promoted priorities are scheduled using preemptive fixed priority (FP) scheduling algorithm. Experimental results demonstrate that this approach reduces the preemption and migration overheads by as much as 90%. Moreover we found that many FP-unschedulable task sets are PNPDP-schedulable.KeywordsDual Priority SchedulingMultiprocessor SchedulingProcrastinationDelayed preemptionCooperative Scheduling