Grid Replicated Storage for the Minimum Intrusion Grid

Masters defence by Jan Wieberg 

For years now the tendency has been that for data to become ever more vital to our lives and our work. Persistent storage devices have vastly increased in capacity but technologies that ensure data safety and no downtime are not yet as ubiquitous. The Minimum intrusion Grid (MiG) is a free and open source grid platform and it is an important element in the computing infrastructure for many of its users. So far, MiG has been restricted to operate on one host machine, with no data-sharing to other MiG servers or redundancy in case of a complete failure of either hardware, software or network at the single server.

In this paper, I present a design and a working implementation of a distributed storage solution built to t the needs of MiG, called Grid Replicated Storage (GRSfs). It is able to oer exible degrees of redundancy and is able to seamlessly fail over to hot-spares if needed. Secondary features include on-demand-fetching from hot-spares and a limited implementation of data branching. I also evaluate the nal results from the prototype, independently as well as in the context of the grid server.

Censor: Josva Kleist, NDGF
Supervisor: Brian Vinter, DIKU