StressCloud

Abstract

The interest in Cloud Computing has grown steadily over recent years, leading to a more intensive use of the datacenters, an increase in their power consumption and a growing complexity of their administration tasks. To assist the administrator in his tasks, the managers of the virtualized datacenters now implement parametrized administration policies, like the Dynamic Power Management in the VMW are software suite. The evaluation of these policies becomes an important and difficult issue for their developers and the datacenters administrator, as the power and performance models of the datacenter’s environment become increasingly complex with the new virtualization techniques. Those techniques become so complex that there is a need for load injection frameworks able to inject resource loads in a tested datacenter instead of model-driven simulators. In this article we present StressCloud, a framework to manipulate the activities of a group of Virtual Machine managed by a Virtualized Data Center Manager and observe the resulting performances. This extensible framework allows to describe, via a scripting language, a scenario of system resources use in Virtual Machines, execute it on a Infrastructure as a Service Provider containing modified Virtual Machines and retrieve the performance data of those Virtual Machines. The resources managed are the CPU, storage and network accesses with extensibility in mind. We show that our load injection model is reliable and the language is expressive enough to describe complex scenarios, with the example of one NASGrid benchmark. We also show that this load injection framework can be used to model hosts’ power consumption based on the hosts’ resources activities.

This work was partially supported by the COST (European Cooperation in Science and Technology) framework, under Action IC0804.