基于推测的无晌应任务自适应容错调度算法-计算机科学.pdfVIP

Vo1. 43 No. 11A 第 43 卷第11A 期 计算机科学 2016 年 11 月 Computer Science Nov 2016 基于推测的无晌应任务自适应容错调度算法 崔云飞吴晓进戴畔程肖郭岗 (北京航天飞行控制中心 北京 100094) 摘 要 已有的基于静态的执行失败判定时间阂值的元响应任务容错调度算法，不能适应大数据处理中心动态的集 群负载。针对该问题，提出判定元响应任务执行失败时间阅值自适应调整方法。基于该模型，设计了自适应的元响应 任务容错调度算法(AFrS) 0 AFrS 算法通过分析作业规模、羊个任务大小和剩余作业推测执行时间等参数，自适应 地调整元响应任务判定执行失败的时间阅值，以减少元响应任务对整体作业执行效率的影响，降低作业响应时间。基 于开发的原型系统，验证了自适应判定方法，测试了算法的性能。实验结果表明，AFrS 算法在作业响应时间等方面 优于已有的无响应任务容错调度算法。 关键词 大数据，容错调度，自适应，推测的，MapReduce 中图法分类号 TP393 文献标识码 A Adaptive Fault-tolerant Scheduling Algorithm for Unresponsive Task ßased on Speculation CUI Yun-fei ^几J Xiacγjin DAI Ye CHENG Xiao GUO Gang (Beijing Aerospace Control Center ,Beijing l00094 ,China) Abstract Current fault-tolerant scheduling algorithm for unresponsive task , based on static execution failed time threshold ,can not adapt to dynamic cluster load of large data processing center. To address this issue ,an adaptive execu tion failed time threshold method was proposed. Based on this method ,an adaptive fault-tolerant scheduling algorithm (AFrS) for unresponsive task was designed. AFrS adjusts unresponsive task s time threshold to be determined failure dynamically and to reduce the job response time ,according to the information of job size ,the size of individual tasks and the remaining operating time. A prototype system using AFrS is developed ,on which the validation of the adaptive exe cution failed time threshold method and the evaluation of AFrS s performance are carried out. It is shown that AFrS outperforms current fault-tolerant scheduling algorithm in term of the job response time. Keywords Big data ,Fault-tolerant sche