Using I/O Schedulers to Reduce I/O Load in Virtualization Environments

Server virtualization helps us system and network administrators manage computer resources and reduce power consumption. However, to make efficient use of virtualization, its various loads, especially I/O operations, require careful consideration. Here we conducted three experiments to examine the effects of I/O load in a virtualization environment. First, We measured the performance of I/O operations executed on a virtual machine (VM) and the power consumption of the host machine, varying the I/O schedulers on both machines. The results varied according to the combinations of the I/O schedulers. Four of the 16 combinations in particular showed low I/O performance and high power consumption. We then investigated the effect of I/O priority to VM processes with a completely fair queuing scheduler on I/O performance. In this case, the I/O execution of high-priority required less than half the time of that of a lower priority process. Finally, we investigated the effects of I/O priority while jointly varying the block size of I/O requests issued by the VMs. I/O execution was effective when the block size was large but not when it was small.