A Graph-Partition-Based Scheduling Policy for Heterogeneous Architectures

Abstract: In order to improve system performance efficiently, a number of systems choose to equip multi-core and many-core processors (such as GPUs). Due to their discrete memory these heterogeneous architectures comprise a distributed system within a computer. A data-flow programming model is attractive in this setting for its ease of expressing concurrency. Programmers only need to define task dependencies without considering how to schedule them on the hardware. However, mapping the resulting task graph onto hardware efficiently remains a challenge. In this paper, we propose a graph-partition scheduling policy for mapping data-flow workloads to heterogeneous hardware. According to our experiments, our graph-partition-based scheduling achieves comparable performance to conventional queue-base approaches.