Preference-oriented partitioning for multiprocessor real-time systems

Abstract

Recently, real-time application models where tasks may have as early as possible (ASAP) or as late as possible (ALAP) execution preferences, while meeting their deadlines, have been proposed and studied. In this work we consider the preference-oriented (PO) real-time scheduling problem for multiprocessor systems. Specifically, we focus on partitioned scheduling of fixed-priority preference-oriented real-time tasks on multiprocessor platforms. Firstly, we explore the use of the Reverse-Preference Priority Assignment (RPPA) scheme, where ALAP tasks are assigned higher priority compared to ASAP tasks on a given processor. Counter-intuitively, this helps the tasks to better fulfill their execution preferences when deployed in the context of the Preference-Oriented Fixed-Priority (POFP) scheduler. Then, considering the complementary execution preferences of the ASAP and ALAP tasks, we propose a preference-oriented partitioning algorithm to allocate tasks to processors. Finally, we extend the algorithm to exploit the period information about the ALAP and ASAP tasks when making the task allocation decisions. The proposed RPPA and preference-oriented partitioning algorithms are evaluated through extensive simulations. The results show that, when the POFP scheduler is adopted, RPPA can significantly improve the execution preferences of ALAP tasks with only marginal impact on ASAP tasks compared to the state-of-the-art priority assignment schemes. Furthermore, compared to the classical utilization based worst-fit-decreasing (WFD) partitioning scheme, the proposed PO partitioning schemes provide more opportunities for both ASAP and ALAP tasks to better fulfill their execution preferences, especially when combined with RPPA and POFP scheduling on each processor.