【摘要】 在大规模的数据中心中,I/O型作业和CPU型作业通常会同时在同一集群上运行,他们之间所需要的资源是互补的.现有的Hadoop调度算法并没有对这两种工作的并行化进行研究,针对这点不足,提出了基于资源感知的调度算法,对作业负载和节点负载进行合理的分类和调度,解决因为资源竞争和不合理调度而影响处理速率的问题.对该算法进行实验及性能分析,在验证了算法在作业完成时间和系统吞吐量方面有很大改进.更多还原

【Abstract】 In a practical large data center,it is a common scenario that I/O-bound jobs and CPU-bound jobs,that demand complementary resources,run simultaneously on the same cluster.In current Hadoop scheduling algorithm,parallelization of these two kinds of job has not been concerned.In this paper,we propose new resource aware scheduling algorithm.This algorithm to classify the type of work and node workload reasonably,and the algorithm mainly solve the problems that affect the processing rate which arise during resource contention.Finally,the experiments and performance analysis of the algorithm proved that the algorithm could significantly improve the system’s throughput.更多还原