Characterizing SMP Virtual Machine Scheduling in Virtualization Environment

Virtualization allows multiple virtual machines to run concurrently on a server. The key to making it possible is to schedule virtual CPUs according to their proportional shares and needs. This paper explores SMP VM scheduling in virtualization environment and seeks to understand its implication to application performance and administration. We designed a monitoring tool in Xen hyper visor to online capture VCPU scheduling events and derive the characteristics. A variety of workloads are measured to analyze the Credit scheduler in three categories: basic scheduling, I/O performance, and co-scheduling. We also discussed how to use our characterizing tool to tune performance for ad hoc applications.