摘要
针对Hadoop平台上调度算法存在的不足,提出了一种改进的调度算法———Triple-Queue算法。在充分考虑数据的本地性后,Triple-Queue算法设计了一种改进的优先级计算模型,以有效地区分用户作业的等级,同时又保证一定程度的公平性,进而减小作业执行时间,避免系统资源浪费。实验结果表明,随着数据量的提高,该算法执行效率明显提高,同时能够较好地解决数据本地性问题。
For solving the shortage of scheduling algorithms on the Hadoop platform,the paper proposed an improved scheduling algorithm-Triple-Queue algorithm.After taking full account of data locality,the Triple-Queue algorithm designs a improved computational model of priority,which can distinguish the user＇s job levels clearly and ensure a certain degree of fairness,so as to reduce the job execution time and avoid wasting system resources.The results of experiment show that the algorithm improves the efficiency significantly and solves the problem of data locality better with the increased amount of data.
出处
《计算机科学》
CSCD
北大核心
2011年第B10期253-256,共4页
Computer Science
