PetaFLOPS Architecture WorkShop (PAWS '96)

April 22 - 25, 1996
Oxnard, California, USA

Objective

PAWS '96 was a federally sponsored workshop that identified opportunities and directions for future research in very high-performance computing systems structures and technologies. The design of this, and other petaFLOPS related workshops, is the achievement of practical petaFLOPS scale computing in the nearest possible time-frame.

Approach

Sponsored by the Defense Advanced Research Projects Agency (DARPA), Department of Energy (DOE), National Aeronautics and Space Administration (NASA), National Security Agency (NSA), and the National Science Foundation (NSF), the workshop consisted of a series of meetings to conduct a detailed examination and evaluation of a set of proposed architecture concepts for near-petaFLOPS performance. The proposed architectures were derived from those selected as part of the NSF solicitation for point-design studies of very high-performance systems.

Accomplishments

PAWS '96 met its goals of setting initial directions for early inquiry into potential petaFLOPS-scale architecture and identifying key issues and objectives that must be addressed by future advanced computing research programs towards the goal of achieving effective petaFLOPS performance.

Significance

The workshop on PetaFLOPS architectures was rich in understanding and findings related to the advance of very high-performance computing towards the goal of achieving practical and useful systems capable of sustaining petaFLOPS level performance. These findings were in two major areas: the findings drawn from all aspects of the workshop deliberations, and a much larger list of supporting detailed findings organized by subdomain of architecture, technology, applications/algorithms, and system software.

Status/Plans

The PetaFLOPS Architecture Workshop identified many critical areas in which advances can and should be made to enhance our understanding of potential approaches and reduce the risk to achieving petaFLOPS computing capability in the earliest possible time-frame. Some immediate actions would be to:

1. Conduct a detailed point design study of COTS-based MIMD petaFLOPS architecture;
2. Perform detailed applications studies of the three archtypical petaFLOPS architecture classes and determine the potential for special purpose structures;
3. Develop latency-tolerant algorithm methods;
4. Design and simulate a hybrid technology architecture incorporating superconducting logic, optical communications, a semiconductor memory hierarchy, and a 3-D holographic storage;
5. Develop an approach for accelerating progress in superconducting memory, submicron fabrication, and interconnect technology.

Points of Contact

Thomas Sterling
Jet Propulsion Laboratory
tron@cacr.caltech.edu
818-354-1793

Michele O'Connell
Boeing Information Services, Inc.
Michele.OConnell@hq.nasa.gov
202-651-8516

Lawrence Picha
Boeing Information Services, Inc.
Larry.Picha@hq.nasa.gov
202-651-8530

Table of Contents | Section Contents -- Basic Research | Subsection Contents -- PetaFLOPS Definition Activities