Bundled Scheduling of Parallel Real-Time Tasks

Abstract

Due to increasing demand for computational power, parallel processing is becoming important for real-time systems. The community has proposed and analyzed several task models and scheduling schemes for parallel applications. Gang scheduling, where all threads of the same task are required to run concurrently, can lead to significant execution time improvements for tightly-synchronized applications. However, existing work on real-time gang scheduling for the rigid model assumes that the number of threads of a parallel task remains constant during its execution, which is not the case for many applications. In this paper, we introduce a novel task model where each task is divided into a sequence of segments called bundles, each requiring a different number of cores, and the threads within each bundle are gang scheduled on an identical multiprocessor. We then derive a sufficient schedulability analysis for sporadic bundled tasks under fixed-priority scheduling. Our evaluation reveals that the proposed model can significantly improve schedulability compared to the rigid model while being competitive with non-gang global thread scheduling and federated scheduling.