Internet has taken over from HPC as "hot technology"

Mannheim, 21-6-97 Internet is hot. It is getting all tthe attention, and even more: there are no start-up companies in high-performance computing (HPC) in the US anymore. HPC does not attract venture capital and there are no new federal megaprograms. Bright graduates all go to Internet and not HPC. The US supercomputing world has to live with that. The same holds for Europe which also has to adapt. Horst Simon, director of the energy supercomputing center NERSC told this to an audience of supercomputer specialists and managers at the Mannheim Supercomputer Conference.

Horst Simon sees 1994 as a milestone in HPC history. A major revolution lead to a market alignment. Major parallel companies like TMC and KSR went out of business and SGI saw the successes of its SMP approach. After that year, concentration went on with HP buying Convex and SGI buying Cray.

The 1997 TOP500 list is completely taken over by CMOS processors. The old, fast but expensive ECL has gone.

In the US, the 1992-1997 years were, according to Simon, those of the HPCC initative. HPC was hot. The strategic goals of the HPCC initative were to extend US technological leadership in HPC; provide dissemination of HPC technology in economy, security, education and environment; spur gains in US productivity. The programme was funded yearly at a $800 million level and coordinated at the highest US governmental level.

"HPCC has accomplished all of its strategic goals", said Simon, "US leadership is firmly established." This can be illustrated in with the TOP500 list. All European vendors (but one) are gone, the Japanes vendors have less than 20% of that market. Clearly, Europe did not find the right answer to the governmental HPCC programme: here, only the users/buyers of supercomputers still represent a strong force .

"HPC has made firm inroads into the commercial world, especially MPP for database applications." stated Simon," A new scientific culture has been established, with large scale collaborations in computational science and a new generation of computationally minded scientists.

Simon noted that the US companies that dominate the HPC scene, are now vertically integrated: all their products are based on workstation technology. The HPC revenues represent only a very small part of the companies' revenues with the exception of SGI/Cray.

"As a result, " Simon said, " HPC customers must get used to a new role: they are not longer at the centre of the attention. Companies must have a commitment to technology, and understand the potential of technology leverage from the high-end, in order to remain in the HPC business. Fortunately for us, the HPC users, they all do understand that (for now)."

The supercomputer centres in the US must get a new role according to Simon. The new model is that they provide intellectual services and are a major facility. Supercomputing hardware is necessary, but not sufficient. New algorithms and strategies must be developed in medium and long-term collaborations with the scientific user community. The New Centre is the working interface between computer science and physical science.

In the US this leads to consolodation of all the NASA syupercomputing operations into one activity, called COSMO. The Department of Defense modernization programme consolidates supercomputing in four production centres: WES and NAVO MIssisipi, Wright-Patterson, Ohio, and ARL, Maryland. as well as two research centres in Maui and Minnesota.

The NSF will only support two centres: NCSA and San Diego. One of the main reasons that NCSA was chosen, according to Simon, is because they showed innovative power with the wide spread used NCSA-Telnet and Mosaic applications, which played an important role in the taking off of the Internet.

The Department of Energy, moved their NERSC supercomputing centre, of which Simon is director, from Livermore to Berkely, resulting in a 20% reduction in budget.

The smaller number of larger centres with intellectual leverage from universities are all based on highly parallel technology. Maybe this model can also be used in Europe suggested Simon.

The impact of the Internet on HPC is large: it attracts all the attention as a new, hot technology. Students do not turn to HPC any more, venture capatilist do no invest in HPC anymore and federal programmes focus on the Internet. There is a next-generation internet initative (NGI) , a $100 million a year presidential initiative, which focusses on high performance networks, advanced network services and new applications like health care and education. The Internet2 initative is a cooperation between 109 member universities. It focuses on research and higher education.

The US HPC programme now in execution is the ASCI initiative on which we publish elsewhere in this issue. Simon doubts whether the $240 million a year put into ASCI will be sufficient to reach the 100 Tflop/s goal in 2002. Since 1993, the high-end HPC market is rather stable. Simon deducted this from an analysis of the value of all the machines in the TOP500. Over the years, this represents a value of US$ 4 billion despite the enormous growth in computing power installed.

Simon concluded with some challenging questions: Which HPC company will be bought by Microsoft? Is SGI/Cray too small to survive? Will Intel re-enter the HPC market? Will there be strategic alliances between Japanese and US vendors?

For the coming years, he expects a predictable but rapid growth in HPC, and although it may not be the hottest thing on Earth, HPC still has a value of its own which sets it apart from other computing areas.

Ad Emmen