I/O-intensive scheduling in multiprocessor virtualized system

Abstract

Virtualization is becoming an important part of cloud computing and high performance calculation. In the virtualized system, the scheduler plays an important role in effecting the performance of whole system. Traditional scheduler that focus on the fairness of VMs would cause problems in I/O performance and latency. In order to eliminate the I/O latency issue, we propose a virtual machine scheduling model based on multiprocessor system. We also implement a prototype in Xen 4.3.0 and evaluate it with several benchmarks. Experiment results demonstrate that our scheduling model can improve the I/O performance effectively. The bandwidth of disk increase by 53.6% and that of network increase by 39.4%. Meanwhile, our method does not change the scheduling algorithm of CPU-intensive VM so that the scheduling fairness is ensured.