Abstract: A History-based Auto-tuning Priority Scheduler（HAPS） is proposed to improve the performance of dealing with hard real-time jobs in Hadoop heterogeneous environments. The algorithm can monitor the real-time progress of jobs, predict the job progress rate by exponential smoothing prediction, calculate the remaining execution time of the job, and estimate the spare time of the job dynamically. According to the spare time, HAPS can update the job priority of job queue in real-time and prefer scheduling the less spare time job. Experimental results show that HAPS can effectively improve the success ratio of dealing with hard real-time jobs in heterogeneous environments.

Key words: scheduling, real-time jobs, exponential smoothing, heterogeneous environments, MapReduce

摘要： 为提升Hadoop集群在异构环境下处理硬实时作业的性能，提出一种基于历史进度自动调整作业优先级的调度算法（HAPS）。该算法实时监控作业进度信息，对作业进度率进行指数平滑预测，计算作业剩余执行时间，动态估算作业空闲时间。并据此实时更新作业队列中作业的优先级顺序，优先调度空闲时间小的作业。实验结果表明，HAPS有效地提高了异构环境下硬实时作业的执行成功率。

关键词: 调度, 实时作业, 指数平滑, 异构环境, MapReduce