会议论文

A Combined Prediction Scheme for the Runtime of tasks in Computing cluster

中文

2014

746-754

2014全国高性能计算学术年会

2014全国高性能计算学术年会论文集

2014-11-06

广州

中国计算机学会

本校为第一机构

复杂的或动态的高性能计算环境需要准确预测任务的运行时间,用于协助调度策略或对资源分配进行未来规划.然而,现有的预测策略的特点决定了唯一的策略并不适用于所有的异构任务.本文提出了一种针对本地任务和远程任务运行时间的组合预测方案(Combined Prediction Scheme,CPS)和预测精度保证(Prediction Accuracy Assurance,PAA)的概念,并把PAA 作为定量评价由某一特定预测策略提供的预测运行时间精度的标准.分析结果表明,CPS具有现行预测策略的优势,PAA能从综合策略所提供的结果中选择接近最优值的预测.CPS增强了计算环境中本地任务和远程任务运行时间的预测精度保证.

　　The complex or dynamic high performance computing environment need to predict the runtime of tasks,in order to assist scheduling strategies or allocation of resources.However,the characteristics of the existing prediction strategies determine that the only policy does not apply to all the heterogeneous tasks.This paper proposes a combined prediction scheme(CPS)for the runtime of local and remote tasks,and a concept of prediction accuracy assurance(PAA),which is a quantitative evaluation standard of the prediction runtime provided by a speci...