Real time scheduler for multiprocessor systems based on continuous control using Timed Continuous Petri Nets

Abstract

Abstract This work exploits Timed Continuous Petri Nets (TCPN) to design and test a novel energy-efficient thermal-aware real-time global scheduler for a hard real-time (HRT) task set running on a multiprocessor system. The TCPN model encompasses both the system and task set, including thermal features. In previous work we calculated the share of each task that must be executed per time interval by solving off-line an Integer Programming Problem Problem (ILP). A subsequent on-line stage allocated jobs to processors. We now perform the allocation off-line too, including an allocation controller and an execution controller in the on-line stage. This adds robustness by ensuring that actual task allocation and execution honor the safe schedule provided off-line. Last, the on-line controllers allow the design of an improved soft RT aperiodic task manager. Also, we experimentally prove that our scheduler yields fewer context switches and migrations on the HRT task set than RUN, a reference algorithm.