National supercomputing policy
Some history, some present, some future
In the Netherlands national supercomputing policy started in 1985, after the Dutch Government published an official Governmental standpoint on supercomputing late 1984. First a working group was established to sort out the various aspects of supercomputing, systems, operational aspects, scientific and industrial use. In 1990 The Netherlands National Computing Facilities Foundation (NCF) was founded by NWO to cover high-end computing matters at a national level. That is beyond the scope and affordability of a single university or institute.
The national policy has developed with technological developments and scientific growth of the importance of Computational Science.
Supercomputing started with single processor vector computers, capable of performing up to 100 Mflop/s. The Netherlands started with a Cray 1S at Shell and a CDC Cyber 205 at SARA. The first supercomputer funded by NWO was a 4-processor Cray Y-MP. Soon the first multi-processor systems came on the market. They shared a single memory across 2, 4, 8 and later up to 16 processors. The next system in the Netherlands was another Cray C90, which started with four (1994) and ended with 12 processors (1999). Processor speed had grown to 333 Mflop/s for the Y-MP and 1000 Mflop/s (1 Gflop/s) for the C90.
Meanwhile the era of massively parallel processing with scalar (instead of vector) processors had started, reaching a level of maturity at standard supercomputing quality.
In 2000 the Cray C90 was followed by a SGI Origin 3800 of 1024 processors, capable of 1 Gflop/s each. So the system reached the level of 1 Tflop/s in total and was theoretically also able to deliver that speed for single applications. That system was tripled in total performance in 2003.
NCF's policy is to have a new system approximately every six years with a significant mid-life upgrade after three years. This schedule serves the scientific community best: the highest performance, the least number of "changes", best affordability, always on track and a kind of predictability for scientists who continuously invest in their software developments.
Science can't live by supercomputers alone, even the science that is called computational e-science. Today any computing resource needs to be embedded in a total environment encompassing high bandwidth low latency wide area networks, like SURFnet, and other facilities, such as large clusters, large data storage areas, visualisation and so on. Therefore NCF's policy focuses on a single integrated grid environment in which all resources can be optimally addressed and used by any researcher that might need them and giving added value to each and every resource within that environment.
NCF cooperates internationally with sister organisations abroad, mainly Europe. These cooperations focus both on high-end supercomputing (HET, DEISA) and grids (e-IRG, EGEE, Sirene)
The future of Dutch supercomputing depends on the Dutch ambitions, which need to be commonly shared among all stakeholders. On behalf of and in the interest of our scientists, the development of our national research centres and the economy at large, it is NCF's aim to steadily be Europe's number five in supercomputer services and resources.