NOMeS: Near-Optimal Metaheuristic Scheduling for MPSoCs

The task scheduling problem for MultiprocessorSystem-on-Chips (MPSoC), which plays a vital role in performance, is an NP-hardproblem. Exploring the whole search space in order to find the optimal solutionis not time efficient, thus metaheuristics are mostly used to find anear-optimal solution in a reasonable amount of time. We propose a novel metaheuristicmethod for near-optimal scheduling that can provide performance guarantees formultiple applications implemented on a shared platform. Applications arerepresented as directed acyclic task graphs (DAG) and are executed on an MPSoCplatform with given communication costs. We introduce a novel multi-populationmethod inspired by both genetic and imperialist competitive algorithms. It is specializedfor the scheduling problem with the goal to improve the convergence policy andselection pressure. The potential of the approach is demonstrated byexperiments using a Sobel filter, a SUSAN filter, RASTA-PLP and JPEG encoder asreal-world case studies.