Towards the Combination of Work-Stealing and Semi-Partitioned Scheduling for Parallel Tasks

We consider the scheduling of sequential and fork-join hard real-time tasks by following semi-partitioned scheduling. In this paper we briefly introduce a modification to the limited migrative model in order to support parallel tasks. Towards the Combination of Work-Stealing and Semi-Partitioned Scheduling for Parallel Tasks Claudio Maia, Luis Nogueira, Patrick Meumeu Yomsi, Luis Miguel Pinho CISTER/INESC-TEC, ISEP, Polytechnic Institute of Porto, Portugal {crrm, lmn, pamyo, lmp}@isep.ipp.pt 1 Motivation and Problem Executing parallel tasks on multicore platforms brings several challenges into the real-time systems domain. Nevertheless, several scheduling approaches have already been proposed in the literature specifically considering this type of tasks. Among the proposed approaches, some use decompositionbased techniques where parallel tasks are treated as a set of sequential tasks so that schedulability analyses developed for sequential tasks can be applied. Others, perform the schedulability by using response-time analysis techniques, or by analysing the resource augmentation bounds of the underlying partitioned or global scheduler. An alternative class of schedulers which has proven to be viable and that has not yet been explored in the context of parallel tasks is that of semipartitioned schedulers. In this class, a subset of tasks is fully partitioned to the cores at design time (i.e., no migration is allowed for these tasks at runtime) and the remaining tasks are scheduled using a global scheduling approach.