Computer Performance Engineering

Abstract

The problem of maximizing the profit achieved by hiring servers from a Cloud and offering virtual machines to paying customers is examined. A number of VMs, each running a user job, can share a server. Hiring a server incurs an initial set-up cost, as well as running costs proportional to the duration of hire. New jobs that cannot start immediately may be lost, or they may be queued. It may or may not be possible to move running VMs from server to server. The effect of these different conditions on several hiring policies, both static and dynamic, is analyzed and evaluated.