Reducing the Number of Context Switches in Real Time Systems

Abstract

In real time scheduling, preemption is one of the causes of run time overhead and large memory requirements. This paper focuses on reducing the number of preemptions in Earliest Deadline First (EDF) scheduling using a technique called Dynamic Preemption Threshold Scheduling (DPTS) in uniprocessor platform. This method is an improvement over existing threshold algorithms, but the complexity is slightly higher. The simulation results show that context switches are reduced by about 91% on an average. This technique is also applied for scheduling of sporadic requests along with periodic tasks. Preemptions that occur when tasks share resources and are required to synchronize are also reduced in this work. Our work also focuses on task set generation with limited hyperperiod (L.C.M. of periods of the tasks).