Upstream: Exposing Performance Information from Cloud Providers to Tenants

Abstract

Infrastructure-as-a-Service (IaaS) is a widely adopted cloud computing paradigm due to its flexibility and competitive prices. To improve resource efficiency, most IaaS providers consolidate several tenants in the same virtualization server, which usually incurs variable performance experiences. In this paper, we evaluate the CPU time received by tenants’ virtual machines (VMs). We present a model that estimates the probability of a VM to receive, at least, a determined fraction of CPU time using limited information about the host and VMs running on it. We constructed this model using a series of experiments with different numbers of CPU cores and co-located VMs.