高能物理计算环境中KVM虚拟机的性能优化与应用

发布时间：2018-01-07 10:07

本文关键词：高能物理计算环境中KVM虚拟机的性能优化与应用　出处：《计算机科学》2015年01期 　论文类型：期刊论文

更多相关文章： 高性能计算 KVM CPU亲和性 扩展页表

【摘要】：高能物理是典型的高性能计算的应用,对CPU计算能力要求很高,并且CPU利用率的高低直接影响高能物理的计算效率。虚拟化技术在实现资源共享和资源高利用率方面表现出很大的优势。基于KVM(Kernel-based Virtual Machine)虚拟机进行性能测试和性能优化。首先对KVM虚拟机的处理器、磁盘IO和网络IO等参数进行测试,给出虚拟机和物理机的性能差异和定量分析,然后从KVM虚拟机架构上分析影响KVM性能的各种因素,从硬件级、内核级对影响性能的因素包括扩展页表EPT(Extented Page Table)和CPU的亲和性(CPU affinity)展开研究,以对KVM进行性能优化。优化结果表明,KVM的CPU性能的损失率可以降低至3%左右。最后,给出了高能物理计算的虚拟集群,结果显示虚拟机群的计算性能能够满足高能物理计算的需求。
[Abstract]:High energy physics (HEP) is a typical application of high performance computing, which requires high CPU computing capability. Virtualization technology has a great advantage in realizing resource sharing and high resource utilization. The performance test and performance optimization of the Kernel-based Virtual Machine are carried out. Firstly, the processor of the KVM Virtual Machine is given. Some parameters such as disk IO and network IO are tested, and the performance difference and quantitative analysis of virtual machine and physical machine are given. Then, the factors influencing the performance of KVM are analyzed from the KVM virtual machine architecture, from the hardware level. At the kernel level, the factors that affect performance include the extended page table EPT(Extented Page table) and the compatibility of CPU CPU affinities). In order to optimize the performance of KVM, the optimization results show that the loss rate of CPU performance can be reduced to about 3%. Finally, the virtual cluster of high energy physics computing is given. The results show that the performance of virtual machine group can meet the demand of high energy physics computing.
【作者单位】：中国科学院高能物理研究所计算中心;
【基金】：国家自然科学基金(11305192,11205179)资助
【分类号】：TP302
【正文快照】： 1引言我国高能物理领域的重大科学工程包括北京谱仪(BE-SIII)[8]、大亚湾反应堆中微子实验[9]、硬X射线调制望远镜(HXMT)、中国散裂中子源等。这些大规模高能物理实验的发展及新发现,离不开对海量数据的处理与分析。因此,如何高效率、高精度地分析海量数据是高能物理计算环境

【相似文献】