Evaluating I/O scheduling in virtual machines based on application load

Abstract

In recent years, cloud computing services and virtualization technology have been widely used. Virtualization requires the access to underlying resources to go through a virtualization layer, which reduces the operation efficiency, especially the access to disk I/O will easily become the bottleneck of the whole system. Therefore, how to improve the I/O performance of virtualization applications has become a hot spot in current researches, especially on I/O scheduling algorithm. While the design and selection of traditional I/O scheduling algorithms are greatly restricted by the seek time and latency of the underlying disks, the virtualization layer in a virtual environment to some extent shields the perception of the scheduling algorithm of virtual machines on the characteristics of the underlying hardware. Whether the traditional algorithms are applicable and how the multi-layer I/O scheduling system in virtualization collaborates to better meet the I/O performance requirements have become pressing issues. In this paper, the authors will explain how the I/O scheduler in Linux system works under different application loads in two scenarios (real machine and virtual machine), and take open-source Xen as examples to test and evaluate the influence of combination of the Dom0 scheduling algorithm and the virtual domain scheduling algorithm on I/O performance under different application loads, and then put forward the preferred proposals of I/O scheduler in virtual domains.