Reclaiming Spare Capacity and Improving Aperiodic Response Times in Real-Time Environments

Abstract

Scheduling recurring task sets that allow some instances of the tasks to be skipped produces holes in the schedule which are nonuniformly distributed. Similarly, when the recurring tasks are not strictly periodic but are sporadic, there is extra processor bandwidth arising because of irregular job arrivals. The additional computation capacity that results from skips or sporadic tasks can be reclaimed to service aperiodic task requests efficiently and quickly. We present techniques for improving the response times of aperiodic tasks by identifying nonuniformly distributed spare capacity—because of skips or sporadic tasks—in the schedule and adding such extra capacity to the capacity queue of a BASH server. These gaps can account for a significant portion of aperiodic capacity, and their reclamation results in considerable improvement to aperiodic response times. We present two schemes: NCLB-CBS, which performs well in periodic real-time environments with firm tasks, and NCLB-CUS, which can be deployed when the basic task set to schedule is sporadic. Evaluation via simulations and implementation suggests that performance improvements for aperiodic tasks can be obtained with limited additional overhead.