Friedel Hossfeld
Central Institute for Applied Mathematics
John von Neumann Institute for Computing
Research Centre Juelich
Germany

Feeding Greedies on Meager Roadmaps

Abstract:

For the first time in computing history, in the 90's we were able to establish a well balanced pyramid ranging from local-area and wide-area broadband networks even suitable for metacomputing, via the diversity of workstation and PC platforms in client-server structures supporting cooperative and even realtime computing, up to the layer of medium-sized innovative computer architectures capable to spread computational science and engineering as a key technology over science and industry and also to support methodologically and capacity-wise, finally, the apex of the pyramid, the infrastructure of top-level supercomputers with the mission of an often nation-wide resource. However, already H. H. Goldstine said: "The history of computers is littered with australopithecanes, the deviant apes that anthropologists keep finding: little evolutionary lines that don't lead anywhere", which is especially true for HPC. Also nowadays, long-range planning for supercomputer centres is a tough task which suffers from sudden deaths of innovative product lines and dead-ends in hailed, but finally meager supercomputer roadmaps. Simultaneously, false promises of cheap solutions may seduce funding agencies to stop enhancing supercomputing despite increasing needs of simulations. However, the time span when the performance level of today's Top500 supercomputers will be reached by the as well exponentially growing PC and workstation performance, amounts up to 14 years. This gap offers opportunities for innovative software product development even for countries where native hardware manufacturers are no longer existent who could act as focal points of national software achievements. Definitely, the missing (super)computer hardware industry demands - at least in Germany - supercomputer centres to play the role of crystallization kernels and attractors of competence in computational science and engineering. The John von Neumann Institute for Computing (NIC) as one of the german supercomputer centres – whose primary supercomputing resources (two 512-processor Cray T3E's, 12-CPU T90 and two 16- and 12-CPU J90's) are provided by the Central Institute for Applied Mathematics (ZAM) at the Research Centre Juelich - acts along these strategic lines by offering computing capabilities to scientists and engineers in universities, research institutes, and industry, and by promoting competence and skills in scientific applications, mathematical algorithms, and visualization methods through research projects and educational programmes. Again, however, the collective mission of supercomputer centres to satisfy also over the next decade, in a nation-wide spectrum, the needs of ever expanding challenging applications, requires to dissolve stubborn hazes over the supercomputer roadmaps and to pave reliable paths into the foreseeable future.