异构集群下的任务调度算法研究

Research of task scheduling algorithm on heterogeneous cluster

针对异构集群下高效节能的任务调度算法进行了研究,提出了一种基于复制的任务调度算法,在任务初始分配的基础上,分别从能源感知和性能—能源平衡两个角度考虑任务的复制。建立了由计算和通信造成的能源消耗的数学模型,并进行了大量的实验。实验结果表明,与已有的BEATA算法相比,该算法能明显地减少异构集群处理并行应用的调度长度和能耗。分析结果发现,任务复制的方法在减少调度长度的同时会增加相应的能耗,能同比优化调度长度和能耗的任务调度方法是今后的研究方向。

Research for fast and energy-efficient task scheduling algorithm on heterogeneous cluster, this paper proposed a duplication-based task scheduling algorithm, which based on the initial allocation of tasks, duplicated tasks from two aspects of energy aware and performance-energy balance. It established mathematical models for the energy consumption caused by computing and communication, and did extensive experiments. Results show that compared with an existing algorithm-balanced energy-aware task allocation （BEATA）, this method can significantly reduce the schedule length and energy consumption of processing parallel applications on heterogeneous cluster, and task duplication method can reduce the schedule length but meanwhile will increase energy consumption. Thus, the task scheduling method which can simultaneously reduce schedule length and energy consumption is the direction of future research.