Hadoop云平台下基于LATE改进的推测执行算法  

Improved Speculated Scheduling Algorithm Based on LATE of Hadoop Cloud Platform

作　　者：卢慧[1] 高弘博[1] 张丰满[1] 王梅[1] 翟嘉伊 肖震[2] 

机构地区：[1]成都纺织高等专科学校软件测试中心,四川成都611731 [2]中国建设银行武汉数据中心,湖北武汉430070

出　　处：《成都纺织高等专科学校学报》2016年第2期200-206,共7页Journal of Chengdu Textile College

基　　金：基于云计算的蜀锦蜀绣综合服务平台(2012FZ0081)

摘　　要：为解决Hadoop云平台下推测执行算法的不足,基于LATE算法提出一种新的任务进度计算方法,并且采用更细粒度的方式选择执行备份任务的节点。实验结果表明,该算法能够更精准的选定掉队者任务并选择合适的执行节点,缩短作业的执行时间,提高云平台的效率。In order to improve the limitation of speculated scheduling algorithm of Hadoop cloud platform,a new task process algorithm based on LATE was proposed and notes of backup task were selected by adopting more fine-grained way. The experiment results showed that such algorithm could more accurately select the stragglers and choose proper executing mote,shorten executing time and improve the efficiency of cloud platform.

关 键 词：HADOOP 云平台 LATE算法 掉队者任务 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]