基于应用特征的PaaS弹性资源管理机制  

Application Feature Based Elastic Resource Management Mechanism on PaaS

作　　者：魏豪[1] 周抒睿 张锐[1] 杨挺[1] 王千祥[1] 

机构地区：[1]北京大学信息科学技术学院软件所高可信软件教育部重点实验室,北京100871

出　　处：《计算机学报》2016年第2期223-236,共14页Chinese Journal of Computers

基　　金：国家"八六三"高技术研究发展计划项目基金(2013AA01A213);国家"九七三"重点基础研究发展规划项目基金(2011CB302604);国家自然科学基金(61421091)资助~~

摘　　要：如何在保障服务质量的前提下,尽可能地节省服务器资源,是云计算环境中弹性资源管理面临的主要问题之一.目前关于云计算环境中弹性机制的研究多数集中在IaaS层,且较少考虑不同应用的具体特征.文中提出了一种基于应用特征的PaaS弹性资源管理机制AFERM(Application Feature based Elastic Resource Management mechanism),主要贡献包括:(1)定义了包含资源开销信息与请求率变化信息的应用特征;(2)设计了一种基于应用执行单元的请求分配机制,在请求数量较大时,将它们划分给多个资源开销相对固定的应用执行单元分别处理;(3)基于应用特征将不同类型的应用搭配部署.作者在自主开发的PaaS平台PAE(Peking university Application Engine)上验证了该机制的可行性和有效性.实验证明,AFERM能够在保证服务质量的前提下,有效地减少服务器的使用:与对比策略相比,AFERM平均可以减少28.3%的虚拟机占用,最多可以减少50%.How to save various resources and keep service quality for clients is one of the main challenges faced by the elastic resource management of cloud computing platforms. Existing elastic management approaches focus mainly on IaaS, and consider few of the features of different applications. In this paper, we presented an application feature based elastic resource management mechanism （AFERM） on PanS. The benefits of AFERM include： （1）defining application feature which is composed of changing pattern of client requests and resource consumptions of the application; （2） proposing an execution unit based requests distribution approach, which divides large number of requests and sends them to multiple execution units with fixed resource consumptions; and （3） deploying applications onto virtual machines based on their features. We evaluated the feasi- bility and validity of AFERM on our own PaaS, PAE （Peking university Application Engine）, and found that AFERM can effectively reduce the number of servers, while keeping the service quality. Compared with the exiting approaches, AFERM can reduce 28. 3% of virtual machines on average, with the highest of 50%.

关 键 词：云计算 PAAS 弹性 应用特征 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]