Multiplexing Low and High QoS Workloads in Virtual Environments

Abstract. Virtualization technology has introduced new ways for man-aging IT infrastructure. The flexible deployment of applications through self-contained virtual machine images has removed the barriers for mul-tiplexing, suspending and migrating applications with their entire execu-tion environment, allowing for a more efficient use of the infrastructure. These developments have given rise to an important challenge regarding the optimal scheduling of virtual machine workloads. In this paper, we specifically address the VM scheduling problem in which workloads that require guaranteed levels of CPU performance are mixed with workloads that do not require such guarantees. We introduce a framework to ana-lyze this scheduling problem and evaluate to what extent such mixed ser-vice delivery is beneficial for a provider of virtualized IT infrastructure. Traditionally providers offer IT resources under a guaranteed and fixed performance profile, which can lead to underutilization. The findings of our simulation study show that through proper tuning of a limited set of parameters, the proposed scheduling algorithm allows for a significant increase in utilization without sacrificing on performance dependability