Summary of Branscomb report

This report presented in August 1993 addresses an opportunity to accelerate the progress in almost every branch of science and engineering by dealing effectively with the development of high performance computing (HPC) resources and their availability to all kinds of organizations in the nation, viz. educational, governmental and industrial. The panel feels that HPC is the key to the nation's technological and economical success. It considers four main challenges that that need to be overcome and suggests suitable remedies for each of them.

1.

Removal of technological barriers relating to the development of suitable parallel hardware, and implementation barriers relating to development of suitable algorithms (i.e. software) to take advantage of this kind of hardware. Also, National Science Foundation (NSF) need to weave together existing work in various developing scientific disciplines, as well as fostering new bridges between pure, applied and computational techniques.

2.

Appropriate acquisition and distribution of HPC resources among the scientists and engineers depending on their need. Each level of institution foreseeing the need for HPC facilities should have atleast a minimum set amount of resources available locally to support their scientists and engineers.

3.

Encouragement of usage of these HPC resources on a broad basis by expanding education and training and encouraging participation by other HPC institutions. This includes providing the right incentive for all three levels of the computational institution pyramid by funding appropriate research.

4.

Determination of a suitable role of NSF in creating the intellectual and management leadership for the future of HPC and its impact on other federal agencies. This mainly deals with the budgeting of the available funds by the NSF to promote critical sectors of High Performance Computing and Communications (HPCC) like networking, parallel architecture, etc.

The report describes four sets of interdependent recommendations to deal with these challenges.

  

Figure: HPC pyramid

1.

The report envisions the HPC pyramid environments and the first recommendation suggests the implementation of a balanced HPC pyramid, each element supporting the others resulting in a stable distribution of the resources. The apex of the pyramid (teraflop class computers) is the need for national capability at the highest level of computing power to deal with ``Grand Challenge'' problems posed to us. The United States should lead the world in HPC by collaborating with big agencies like ARPA and DoE to expand access to all levels of the pyramid for all sectors of the nation. It is also aimed that the research universities should be assisted to acquire mid-range machines over the next five years, and that that NSF should double the current annual level of investment in scientific and engineering workstations for its 20,000 principal investigators. It was also recommended that NSF support the development of new parallel configurations for HPC applications.

2.

To accelerate progress in developing the required HPC technology, NSF should create a challenge program in computer science with grant size and equipment access to support the systems and algorithm research. NSF should focus attention on support for the design and development of computational techniques, algorithmic methodology, and mathematical, physical and engineering models to make efficient use of the machines. NSF should also set up an agency-wide task force to develop a way to remove the imbalance in the HPC pyramid caused by the the under-investment in the emerging mid-range scalable and parallel computers and the inequality of access to stand-alone potentially networked workstations.

3.

The Supercomputer Centers should be retained and their missions, as they have evolved, should be reaffirmed. An open re-competition of the four Supercomputer Centers should be avoided at this time, instead favoring periodic performance evaluation and competition for some elements of their activities. The NSF should continue to provide funding to support the Supercomputer Centers' HPC capacity. The NSF should review the administrative procedures used to allocate center resources to ensure that the burden on scientists applying for research support is minimized. The NSF should give strong emphasis to its education mission in HPC, and should actively seek collaboration with state-sponsored and other HPC centers not supported primarily on NSF funding.

4.

The National Science Board should urge the Office of Science and Technology Policy (OSTP) to establish an advisory committee representing the states, HPC users, NSF Supercomputer Centers, computer manufacturers, computer and computational scientists which should report to High Performance Computing, Communications, and Information Technology (HPCCIT) subcommittee.