###
Journal of Software:2015.26(8):2074-2090

一种基于硬件计数器的虚拟机性能干扰估算方法
王卅,张文博,吴恒,宋云奎,魏峻,钟华,黄涛
(中国科学院 软件研究所 软件工程技术中心, 北京 100190;计算机科学国家重点实验室(中国科学院 软件研究所), 北京 100190;中国科学院大学, 北京 100049)
Approach of Quantifying Virtual Machine Performance Interference Based on Hardware Performance Counter
WANG Sa,ZHANG Wen-Bo,WU Heng,SONG Yun-Kui,WEI Jun,ZHONG Hua,HUANG Tao
(Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China;State Key Laboratory of Computer Science (Institute of Software, The Chinese Academy of Sciences), Beijing 100190, China;University of Chinese Academy of Sciences, Beijing 100049, China)
Abstract
Chart / table
Reference
Similar Articles
Article :Browse 1818   Download 1897
Received:April 28, 2014    Revised:August 24, 2014
> 中文摘要: 虚拟化技术已成为云计算平台中的关键性支撑技术.它极大地提高了数据中心的资源利用率,降低了管理成本和能源消耗,但同时也为数据中心带来了新的问题——性能干扰.同一平台上的多虚拟机过度竞争某一底层硬件资源(如CPU,Cache等),会造成虚拟机性能严重下降;而出于安全性和可移植性的考虑,底层平台管理者需要尽量避免侵入式监测上层虚拟机,因而,如何透明而有效地从底层估算虚拟机性能干扰,成为虚拟化平台管理者必须面临的一个挑战.为应对以上挑战,提出了一种基于硬件计数器的虚拟机性能干扰估算方法.硬件计数器是程序运行期间产生的硬件事件信息(如CPU时间片、缓存失效次数等),已有工作主要利用大规模分布式系统任务相似性查找产生异常硬件计数器数据的节点,而没有探究硬件事件变化与性能干扰之间的直接关系.通过实验研究发现,硬件计数器(last level cache misses rates,简称LLC misses rates)与不同资源需求的应用性能干扰存在不同的关联关系;以此建立虚拟机性能干扰估算模型,估算虚拟机性能.实验结果表明:该方法可以有效地预测CPU密集型应用和网络密集型应用的性能干扰大小,并仅为系统带来小于10%的开销.
Abstract:In IaaS platforms, hardware infrastructures are sliced into multiple virtual machines (VMs) to provide computing capabilities for users. Virtualization greatly improves the resource utilization, however it introduces potential risk of variation in VM performance. VMs co-located together have a high probability of performance degradation when one of the VMs behaves as a noisy neighbor competing hardware resource with other victims. How to efficiently monitor and quantify this type of performance interference thus becomes a key challenge for IaaS providers. To address these challenges, this study presents an approach which transparently monitors and quantifies VMs interferences through low-level metrics with hardware performance counters (HPCs). The approach explores the information within HPC and LLC miss rates, builds performance prediction model and quantifies performance interference of different (CPU-bound and net-bound) VMs. Experimental results show that the proposed approach can predict the performance degradation effectively with an acceptable overhead that is lower than 10%.
文章编号:     中图分类号:    文献标志码:
基金项目:国家自然科学基金(61173003); 国家高技术研究发展计划(863)(2012AA011204) 国家自然科学基金(61173003); 国家高技术研究发展计划(863)(2012AA011204)
Foundation items:
Author NameAffiliationE-mail
WANG Sa Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China
State Key Laboratory of Computer Science (Institute of Software, The Chinese Academy of Sciences), Beijing 100190, China
University of Chinese Academy of Sciences, Beijing 100049, China 
wangsa09@otcaix.iscas.ac.cn 
ZHANG Wen-Bo Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China  
WU Heng Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China
State Key Laboratory of Computer Science (Institute of Software, The Chinese Academy of Sciences), Beijing 100190, China
University of Chinese Academy of Sciences, Beijing 100049, China 
 
SONG Yun-Kui Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China  
WEI Jun Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China
State Key Laboratory of Computer Science (Institute of Software, The Chinese Academy of Sciences), Beijing 100190, China 
 
ZHONG Hua Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China
State Key Laboratory of Computer Science (Institute of Software, The Chinese Academy of Sciences), Beijing 100190, China 
 
HUANG Tao Technology Center of Software Engineering, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China
State Key Laboratory of Computer Science (Institute of Software, The Chinese Academy of Sciences), Beijing 100190, China 
 
Reference text:

王卅,张文博,吴恒,宋云奎,魏峻,钟华,黄涛.一种基于硬件计数器的虚拟机性能干扰估算方法.软件学报,2015,26(8):2074-2090

WANG Sa,ZHANG Wen-Bo,WU Heng,SONG Yun-Kui,WEI Jun,ZHONG Hua,HUANG Tao.Approach of Quantifying Virtual Machine Performance Interference Based on Hardware Performance Counter.Journal of Software,2015,26(8):2074-2090