While grids are composed of many diverse resources, their applications usually require a very specific, validated environment.
As a result, applications that work on a developer’s desktop may only function “out of the box” on a small fraction of the total number of compute resources potentially available to scientists on the grid. This is one of the primary obstacles users face in grid computing.
A grid of your virtual machines
One solution is to take the developer’s desktop and scale it to hundreds of nodes by mapping the desktop onto hundreds of virtual machines and deploying them onto grid resources.
To facilitate this mode of using the grid, researchers at the Computation Institute at the University of Chicago recently announced the availability of a “science cloud,” codenamed Nimbus. Nimbus is available to scientific projects and comes with a simple client that allows scientists to provision virtual machines using a leasing model similar to Amazon’s EC2 service.
High in the science cloud
The primary objective of Nimbus is to make it easy for scientific and educational projects to experiment with this mode of resource provisioning.
A secondary objective is to learn how the scientific community would like to use a cloud of this type and respond to those requirements by developing features and enhancements to make cloud computing attractive to scientific projects.
In a typical interaction on the Nimbus cloud, a user makes a request to deploy a virtual machine image. Users may use one of the virtual machine images already available on the cloud or upload their own image. On deployment, the image is configured with an ssh public key provided by the user which allows access to the cloud.
The future of Nimbus
Nimbus is a beta project deployed on a modest allocation of resources; however, more providers are interested in exploring this mode of resource provisioning: the University of Florida is expected to provide more resources within the next week and other centers have also expressed interest.
“Provisioning resources via virtual machines has the potential to solve many problems we face today in grid computing,” says Jose Fortes, professor at the University of Florida. “It is important to create conditions in which scientists can experiment with this new mode of resource provisioning and learn of their requirements early, so that we can create the best possible platform for science.”
Jérôme Lauret, software and computing project leader for the Solenoidal TRack At RHIC (STAR), is one of the first to make use of Nimbus and says it has made complex computing much easier.
“Our applications rely on multiple tools and packages that are complex and time-consuming to install,” he says. “Now Nimbus allows the preparation of a single and validated image for use on multiple nodes and multiple sites, reducing the support overhead and making complex tasks such a full event reconstruction easily possible.”