Sandboxes for Grid Computing

PhD-defence by Rasmus Andersen 


This dissertation explores various ways of utilizing sandboxes on computational resources in a Grid system. Sandboxes in a Grid context has many benefits for both users of the Grid and the resource owners: Most prominently, for users, sandboxes can create the illusion of having seamless access to one big supercomputer composed of many heterogeneous and geographically dispersed computers, and for the resource owners, sandboxes provide an extra layer of security, which is very important when allowing execution of arbitrary code on personal machines.

Using existing virtual machines as sandboxes, it is shown how to combine Public Resource Computing and Grid Computing to form a very powerful and easily accessible computing platform. During the course of the study, several projects from other fields of science have been conducted on this platform with promising results.

The main drawback with standard virtual machines is the performance loss incurred by the virtualization layer and the lack of portability. In the second part of the project, this issue is addressed by developing a new high-performance virtual machine designed specifically for scientific applications. By using a platform-independent bytecode, and taking advantage of the nature of scientific applications, the machine is portable across radically different architectures and operates close to native speed.

Assessment Committee:

Chairman: Jon Sporring
Professor Lasse Natvig ( Norwegian University of Technology and Science)
Associate Professor Arne Skou ( Department of Computer Science at Aalborg University)

Academic supervisor: Brian Vinter