Abstract :
Summary form only: Over the last 50 years, computational sciences and engineering have advanced the development of new and ever faster and better computers, algorithms, and software tools, and vice versa, computers have advanced sciences. But the better the technologies the more demanding our computational science applications. This progress is everywhere, at the level of technologies and computer architectures, at the middleware level, at the algorithm and application level, and even at the level of computing paradigms, where we evolved from mainframes, to vector and parallel computers, to grids and clouds, recently. We successfully optimized our algorithms and mapped them to the underlying architecture, e.g. with overlapping communication with computation, better load balancing through domain decomposition into parallel processes, or using library routines optimized for the specific architecture and processors. In our presentation, we will concentrate on the use of grid and cloud technologies for HPC, focusing on the European EU-funded DEISA Distributed European Infrastructure for Supercomputing Applications project, analyzing different HPC loads and their suitability for grids and for clouds, and taking a closer look at lessons learned and recommendations on how to build sustainable e-Infrastructures for computational sciences and engineering in the future.
