Scheduling multiple virtual environments in cloud federations for distributed calculations

Abstract

In the pool of cloud providers that are currently available there is a lack of standardised APIs and

brokering tools to effectively distribute high throughput calculations among them. Moreover, the current

middleware tools are not able to straightforwardly provision the ephemeral and specific environments

that certain codes and simulations require. These facts prevent the massive portability of legacy

applications to cloud environments. Such an issue can be overcome by effectively scheduling the

distributed calculations using the basic capacities offered by cloud federations. In this work, a framework

achieving such a goal is presented: a pilot system (GWpilot) that has been improved with cloud computing

capabilities (GWcloud). This framework profits from the expertise acquired in grid federations and

provides interesting features that make it more efficient, flexible and useable than other approaches. Thus,

decentralisation, middleware independence, dynamic brokering, on-demand provisioning of specific

virtual images, compatibility with legacy applications, and the efficient accomplishment of short tasks,

among other features, are achieved. Not only this, the new framework is multi-user and multi-application,

dynamically instantiating virtual machines depending on the available and demanded resources, i.e.

allowing users to consolidate theirown resource provisioning. Results presented in this work demonstrate

these features by efficiently executing several legacy applications with very different requirements on the

FedCloud infrastructure at the same time.