The eMinerals project was a collaboration between environmental scientists, physicists, chemists and computer scientists from the universities of Bath, Birkbeck College, Cambridge, Reading, UCL, the Royal Institution and the CCLRC Daresbury. One of the aims of the project was to allow scientists and computing experts to jointly produce the grid infrastructure, tools and technologies, that will facilitate the science, which involves simulations of environmental processes at a molecular level with increased levels of realism.
As one of the computer scientists on the team, my tasks have included the building of the eMinerals minigrid, a production-level grid infrastructure incorporating a wide range of dedicated and shared resources. This infrastructure is primarily composed of PBS managed linux clusters, IBM pSeries clusters, the large UCL and Cambridge Condor pools, a sunfire 880 parallel computer and, indirectly, a number of UK e-science level 2 grid resources, such as HPCx. All this is brought together using the Globus toolkit for computational purposes and the Storage Resource Broker (SRB) for data storage, as well as client side tools such as Condor-G/glide-in and a web-based portal currently in development. Our mini-grid also encapsulates a number of collaborative tools and technologies, such as the Personal Interface to the Access Grid (PIG), helpdesk software systems, and application sharing software. Description and examples of science accomplished using our eMinerals minigrid are available [ here ], [ here ] and [ here ].
I was also responsible for the deployment, management and support of the previously mentionned UCL Condor pool, which currently incorporates over 1400 nodes - a figure set to further increase in the coming months. At time of writing, the UCL Condor pool is still the largest academic pool in the UK and since October 2003, has managed, processed and returned over 300 CPU years of computational chemistry, physics and geological programs for users all over the UK academic community. An example of large scale computational science accomplished using our resources is available [ here ]. Further information is available [ here ].
In addition my work on Condor has also involved the study of interoperability between Condor's architecture and emerging grid standards, such as the Open Grid Services architecture [ OGSA ]. Specifically, I have participated in the development of a Web Service interface to Condor services with the Condor team at the university of Wisonsin-Madison. Condor birdbath [ jump ] is now incorporated in versions of Condor 7.5 and above. This has also involved the development of appropriate extensions to the GridSAM (OMII) job submission service to support Condor's WS interface.
My research interests rotate around the development of scheduling strategies in best-effort autonomous grid environments. My PhD thesis involved the definition of an infrastructure which used adaptive and cooperative mechanisms to optimise the selection, handling and coordination of tasks in grid environments. I was specifically looking at the use of prediction mechanisms such as Kalman Filter theory to minimise job throughput [ jump ], as well as reputation based systems to reach a common concensus on scheduling policies for independent resource managers. I was also particularly interested in the use and incorporation of web standards into existing production level tools, such as JSDL, BPEL and WSRF, which introduce new capabilities that I aim to take advantage of in my work [ jump ].
More: [ RESERVOIR ]