ENGINEERING

The power to connect supercomputing scientists working at European and US participating sites so that they can create or access collaborative data through a common, scalable global file system was demonstrated by researchers from both sides of the Atlantic. The demonstration took place at the Supercomputing Conference SC|05 in Seattle, November 2005. It linked DEISA, the Sixth Framework Programme-financed European supercomputing Grid infrastructure project with the corresponding US supercomputing cyber-infrastructure TeraGrid. “DEISA is a Grid highly focused on supercomputing and the needs of the supercomputer user communities,” says DEISA’s Hermann Lederer from the Garching Computing Centre in Germany. “There are only two or perhaps three Grids of this kind in the world, that integrate a relatively small number of leading, high-performance computing platforms.” This sets it apart from larger Grids that connect a number of medium range clusters that can comprise of more than 90 institutions. “But there is still another important difference to most other Grids at the conceptual level,” continues Lederer. “As a new key element, DEISA has introduced the global file system concept among different sites from the beginning in order to reduce the dependence on too many components of complex middleware packages. Teragrid has also adopted this approach, and a fruitful cooperation has started between the two Grid projects.” On show at SC05 was how any scientist, accessing DEISA from any of its sites in France, Germany or Italy, or accessing TeraGrid from any of the US participating sites - The San Diego Supercomputer Center, The University of Chicago, and Indiana University, to directly create or access collaborative data stored in the global file systems of DEISA and TeraGrid with one common file address space. Even more importantly applications which, executed at any of the participating sites, could transparently access data in the common file address space. “This drastically reduces the complexity of measures for data access through geographically distributed sites,” comments Lederer. To illustrate this, the featured applications included a protein structure prediction carried out at the US’ San Diego Supercomputer Centre, a gyrokinetic turbulence simulation at Germany’s Garching Computing Centre of the Max Planck Society and also cosmological simulations from both sites. These single site applications wrote their results to the intercontinental global file system, ready for further processing from other Grid access points. The demonstration at SC05 underlined the importance of interoperability when Grids enter production. As a result during the conference leading Grid projects from all continents held a joint meeting where tackling interoperability was cited as a high priority. The network connection between DEISA and TeraGrid is expected to become persistent at some time in the near future. Both have begun using the high performance wide-area global file system GPFS from IBM. This opens up totally new modes of operation within Grid infrastructures, especially in supercomputing Grids with a fairly limited number of participating sites. DEISA is also currently expanding its global file system to include The Netherlands, Spain and the UK, and in doing so is more tightly integrating other leading European supercomputing centers, aiding the advancement of computational sciences in the area of high-performance computing. DEISA’s next steps include the deployment and operation of a global data management infrastructure for high performance computing across the continent and the deployment of services needed for extreme computing in Europe. There are also plans to upgrade the dedicated network to 10 Gigabits per second and for closer cooperation with other major Grid projects in the world. Thanks to the work of DEISA, such a common data repository with fast access, accessible both by applications running anywhere in the Grid, and by scientists working at any partner site as an entry point to the Grid, will greatly facilitate cooperative work as scientific communities increasingly become distributed geographically.