动态标签匹配DLMS调度器设计与实现  

Design and Implementation of Dynamic Label Matching Scheduler

作　　者：毛韦 竹翠[1] 

机构地区：[1]北京工业大学计算机学院,北京100124

出　　处：《软件导刊》2017年第9期95-99,共5页Software Guide

摘　　要：针对Hadoop集群节点性能差异大、资源分配随机、执行时间过长的问题,提出一种将节点性能标签(简称节点标签)和作业类别标签(简称作业标签)进行动态匹配的调度器。节点初始分类并赋予原始节点标签,节点检测自身性能指标生成动态节点标签,作业根据部分运行信息进行分类并生成作业标签,资源调度器将节点资源分配给对应标签的作业。实验结果表明,相对于YARN中自带的调度器,其在作业执行时间上有很大缩短。In order to solve the problem of big performance difference, random resource allocation, and long time execution in Hadoop cluster nodes, this paper propose a scheduler which allocates the Node Performance Label（NPL） and Job Category Label （JCL） dynamically. The node makes the classification of initialization and is assigned original node label. The node detects its own performance metrics to generate dynamic node labels. The job is classified based on part of the run information to generate the job label. The resource scheduler assigns the node resource to the corresponding label job. The experimental results shows that the scheduler has a certain degree of shorten in the time of job execution compared with one that comes with YARN.

关 键 词：HADOOP 资源调度器 动态匹配 动态标签 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]