异构多核计算系统的Codelet任务调度策略

Codelet数据流计算模型在处理大规模并行计算任务时效果显著，但该模型目前缺少在异构多核环境中的任务调度策略。因此，提出了一种在异构多核环境下基于蚁群算法的Codelet任务调度策略。该调度策略将启发式算法与蚁群算法相融合，在发挥各自优势的同时克服了启发式算法不能得出最优解的缺陷以及蚁群算法初始信息匮乏的问题。实验结果表明，智能蚁群任务调度策略相比Codelet运行时系统中原生的动态调度和静态调度策略具有更高的执行效率。

Task scheduling policy of Codelet on heterogeneous multicore computing system

Codelet dataflow model has significant effects on gaining high performance of computing large-scale parallel tasks, but the model currently lacks scheduling policy in heterogeneous multi-core environment. Regarding to this issue, this paper proposed a Codelet task scheduling strategy by fusing ant colony algorithm with a heuristic approach in heterogeneous multicore environment. It had both advantages of the heuristic algorithm and ant colony algorithm, and it overcame both defects of the heuristic algorithm that could not derive an optimal solution and the defects of the ant colony algorithm that was lack of the initial information. The experimental result shows, the smart ant colony scheduling policy is much more efficient than the native dynamic and static scheduling policies in the runtime system implementation of the Codelet model.