The transition to massively parallel computing within a production environment at a DOE access center
In contemplating the transition from sequential to MP computing, the National Energy Research Supercomputer Center (NERSC) is faced with the frictions inherent in the duality of its mission. There have been two goals, the first has been to provide a stable, serviceable, production environment to the user base, the second to bring the most capable early serial supercomputers to the Center to make possible the leading edge simulations. This seeming conundrum has in reality been a source of strength. The task of meeting both goals was faced before with the CRAY 1 which, as delivered, was all iron; so the problems associated with the advent of parallel computers are not entirely new, but they are serious. Current vector supercomputers, such as the C90, offer mature production environments, including software tools, a large applications base, and generality; these machines can be used to attack the spectrum of scientific applications by a large user base knowledgeable in programming techniques for this architecture. Parallel computers to date have offered less developed, even rudimentary, working environments, a sparse applications base, and forced specialization. They have been specialized in terms of programming models, and specialized in terms of the kinds of applications which would do well on the machines. Given this context, why do many service computer centers feel that now is the time to cease or slow the procurement of traditional vector supercomputers in favor of MP systems? What are some of the issues that NERSC must face to engineer a smooth transition? The answers to these questions are multifaceted and by no means completely clear. However, a route exists as a result of early efforts at the Laboratories combined with research within the HPCC Program. One can begin with an analysis of why the hardware and software appearing shortly should be made available to the mainstream, and then address what would be required in an initial production environment.
- Research Organization:
- Lawrence Livermore National Lab., CA (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 10183045
- Report Number(s):
- UCRL-JC-113508; CONF-9304159-2; ON: DE93019155
- Resource Relation:
- Conference: Conference on high speed computing,Salisham, OR (United States),Apr 1993; Other Information: PBD: Apr 1993
- Country of Publication:
- United States
- Language:
- English
Similar Records
Large Scale Computing and Storage Requirements for Basic Energy Sciences: Target 2017. Report of the NERSC Requirements Review Conducted October 8–9, 2013, Gathersburg, MD
The ASC Sequoia Programming Model