Cong
Guo,
PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Consolidation of multiple workloads is cost-effective for system operators. However, it is difficult to determine how to share resources among multiple tenants to achieve both performance isolation and work conservation. The primary shared resource in the server are the CPU cores. We show that current solutions cannot handle CPU sharing very well in various multi-tenancy scenarios.
We propose a manager system which dynamically adjusts the CPU allocation according to the actual requirements of multiple workloads, and supports different sharing policies. Experimental results in different scenarios show that our manager can achieve both isolation and work-conservation at the same time.