Page 315 - 《软件学报》2020年第12期

软件学学报 ISSN 1000-9825, CODEN RUXUEEW E-mail: jjos@iscas.ac.cn
Journaal of Software,2020,31(12):3981−3999 [[doi: 10.13328/j.cnkki.jos.005892] http:///www.jos.org.cn
©中国科学院软件研究所所版权所有. Tel: +886-10-62562563

∗
一种种云环境中中的动态细粒度资源调调度方法

周墨颂颂, 董小社, 陈 衡, 张兴军

(西安交交通大学 电子与信息工程学院,陕西西 西安 710049))
通讯作作者: 陈衡, E-maill: hengchen@mail..xjtu.edu.cn

摘 要要: 云计算平台台中普遍采用固定定资源量的粗粒度度资源分配方式,由此会引起资源源碎片、过度分配配、低集群资
源利用用率等问题.针对对此问题,提出一种种细粒度资源调调度方法,该方法根根据相似任务运行行时信息推测任任务资源需求;
将任务务划分为若干执执行阶段,分阶段匹匹配资源,从分配配时间和分配资源源量两方面细化资资源分配粒度;资资源匹配过程
中,基基于资源可压缩特特性进一步提高资资源利用率和性能能;采用资源监控控、策略调整、约约束检查等机制保保证资源使用
效率和和负载性能.在开开源云资源管理平平台中,基于细粒粒度资源调度方法法实现了调度器.实实验结果表明:细细粒度资源调
度方法法可以在不丧失公公平性且调度响响应时间可接受的的前提下,细化资源源匹配的粒度,有有效提高云计算平平台资源利用
率和性性能.
关键词词: 细粒度;调度度;云计算;资源管管理;平台优化
中图法法分类号: TP3116

中文引引用格式: 周墨颂颂,董小社,陈衡,张兴兴军.一种云环境中的动态细粒度资资源调度方法.软件件学报,2020,31(122):3981−3999.
http://wwww.jos.org.cn/10000-9825/5892.htmm
英文引引用格式: Zhou MMS, Dong XS, Cheen H, Zhang XJ. DDynamically fine-ggrained schedulingg method in cloudd environment.
Ruan JJian Xue Bao/Journnal of Software, 20020,31(12):3981−33999 (in Chinese). hhttp://www.jos.orgg.cn/1000-9825/58992.htm

Dynaamically Fine--grained Scheeduling Methood in Cloud Environment
ZHOUU Mo-Song, DOONG Xiao-She, CHEN Heng, ZHANG Xing-Jun
(Facultty of Electronics and Information Enginneering, Xi’an Jiaotoong University, Xi’aan 710049, China)
Abstraact: The coarse-grrained scheduling uused in cloud compuuting platform alloccates fixed quantity resources to tasks.. However, this
allocattion can easily lead to problems such aas resource fragmenntation, over-commiitment and inefficieent resource utilizatiion. This study
proposses a dynamically fiine-grained schedulling method to resolve those problems.. This method estimmates resource requiirement of task
accordding to similar taskss and divides tasks into execution stagges according to thee task requirement, and it also matches task resource
requireement and availablee server resources bby stages to refine two aspects of alloocation granularity: allocation duration and allocation
quantitty. Furthermore, thiis method may commpress resource allocation to further immprove resource utillization and performmance, and this
methodd uses several mechhanisms including ruuntime resource moonitoring, allocation policy adjustments, and scheduling coonstraint checks
to ensuure resource utilizattion and performancce of cloud computing platform. Based on this method, a sscheduler has been iimplemented in
the open source cloud commputing platform YYarn. The test results show that the dynnamically fine-grainned scheduling methhod can resolve
resourcce allocation probleems by significantlyy improving resourrce utilization and pperformance with aacceptable fairness and scheduling
responnse times.
Key wwords: fine-grainedd; schedule; cloud computing; resource management; platfoorm optimization

资
资源管理与作业业调度是云计算资资源管理平台中中的关键点之一..随着云计算平台台上负载的多样样性和动态性
[1]
日益增增加,原有的资源源管理与作业调度度方式的有效性性严重降低 .

∗ 基基金项目: 国家重点点研发计划(2016YFFB0200902); 国家自然科学基金(615772394)
FFoundation item: NNational Key Reseaarch and Developmment Program of CChina (2016YFB02000902); National NNatural Science
Foundaation of China (615772394)
收收稿时间: 2017-10--16; 修改时间: 2018-06-20; 采用时间间: 2019-09-24