Many-Core Scheduling of Data Parallel Applications Using SMT Solvers

To program recently developed many-core systems-on-chip two traditionally separate performance optimization problems have to be solved together. Firstly, it is the parallel scheduling on a shared-memory multi-core system. Secondly, it is the co-scheduling of network communication and processor computation. This is because many-core systems are networks of multi-core clusters. In this paper, we demonstrate the applicability of modern constraint solvers to efficiently schedule parallel applications on many-cores and validate the results by running benchmarks on a real many-core platform.