面向网络性能优化的虚拟计算资源调度机制研究  

Study on virtual computing resource scheduling for network performance optimization

作　　者：王煜炜[1,2] 刘敏[1] 房秉毅[3] 秦晨翀 闫小龙[1,2] 

机构地区：[1]中国科学院计算技术研究所,北京100190 [2]中国科学院大学,北京100049 [3]中国联合网络通信集团有限公司,北京100032

出　　处：《通信学报》2016年第8期105-118,共14页Journal on Communications

基　　金：国家自然科学重点基金资助项目(No.61132001;No.61120106008;No.61472402;No.61472404;No.61272474;No.61202410)~~

摘　　要：针对基于Xen的v CPU调度机制对虚拟机网络性能的影响进行了深入研究和分析。提出一种高效、准确、轻量级的网络排队敏感类型虚拟机(NSVM)识别方法,可根据当前虚拟机I/O传输特征将容易受到影响的虚拟机进行准确识别和区分。进而设计一种新型虚拟计算资源调度和分配机制Diff-Scheduler,将不同类型虚拟机的v CPU实施分池隔离调度,同时提高NSVM类型虚拟机v CPU的调度频率。原型系统实验结果表明,相比Xen默认的调度机制,Diff-Scheduler能够大幅提高虚拟机网络性能,同时保证计算资源分配的公平性。A deep insight into the relationship between vCPU scheduling and I/O transmit in Xen was provided. Then an effective and lightweight recognition method, through which could identify the so-called NSVM （network queuing sensi- tive virtual machine） that was more vulnerable to the congestion in I/O transmit was put forward. Furthermore, a novel mechanism for resource assignment and scheduling called Diff-Scheduler was proposed. It could schedule the vCPU of the NSVM more frequently than other VM in different pools independently. Evaluations based on a prototype of Xen platform featured Diff-Scheduler show that the proposed mechanism significantly improves the network performance of VM. Specifically, comparing with the default mechanism of Xen, Diff-Scheduler proposed jointly enhances throughput, latency remarkably and ensures the fairness of resource allocation at the same time.

关 键 词：云计算 虚拟化 v CPU调度 网络排队敏感型 虚拟计算资源分配 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]