Xen的I/O虚拟化性能分析与优化

Performance Analysis and Optimization on I/O Virtualization of Xen

在Credit算法应用中,由I/O事务唤醒的VCPU处于最高优先级BOOST状态,优先抢占PCPU资源,使I/O操作的响应速度提高,但多个虚拟机同时进行I/O操作时,会引起较长延时和公平性原则被破坏问题.针对这个问题,研究分析SEDF算法、Credit算法、Credit2算法,提出L-Credit调度算法解决多个虚拟机同时进行I/O操作引起响应延迟的问题.通过监测I/O设备环共享页面中响应和请求的个数的方法,对处于BOOST状态下I/O操作进一步细化排序,使稀疏型I/O操作较密集型I/O操作先调用执行.通过对L-Credit算法与Credit算法在同一应用场景下反复对比实验,得出L-Credit算法可以提高I/O响应性能,并且继承了Credit算法负载均衡和按比例公平共享的特点.

In the application of Credit algorithms, the VCPU awakened by the I/O transaction is in the highest priority BOOST state, which gives it priority to gain access to the PCPU resources and improves the response speed of the I/O operation, but that will cause long time delay and destroy fairness principle, when multiple virtual machines are doing the operation of I/O at the same time. To solve this problem, SEDF algorithm, Credit algorithm, Credit2 algorithm are analyzed and L-Credit scheduling algorithm is proposed to reduce the response delay caused by multiple virtual machines＇ concurrent I/O operation. By monitoring I/O device ring sharing page to get the number of requests, which can further refine the sort in the I/O state BOOST operation, so that sparse type I/O operation can the implementation before the I/O intensive operations. According to the comparison report of the L-Credit algorithm and Credit algorithm in the same application scenario experiment, L-Credit algorithm can improve the performance of the I/O response, and inherits the Credit algorithm load balance and the characteristics of proportional fair sharing.