December 1, 1995
Jarrett S. Cohen
Goddard Space Flight Center
Greenbelt, Md. 20771
(301-286-2744, jarrett.cohen@gsfc.nasa.gov)
From virtual reality demonstrations to technical papers, NASA's Earth and Space Sciences (ESS) Project is participating in a wide range of activities at Supercomputing '95, December 3 - 8, San Diego, Calif. The Project emphasizes the use of scalable parallel computers for solving Grand Challenge problems. Part of the NASA High Performance Computing and Communications Program, it is based at Goddard Space Flight Center (GSFC) and the Jet Propulsion Laboratory (JPL).
One of the ESS Project's eight Grand Challenge teams will be displaying "4D Navigation of Astrophysical Turbulence Data" in a virtual reality environment; 3D texture mapping will be employed to create a volume rendering scene of supercomputer data obtained from the IBM SP-1 at Argonne National Laboratory (ANL). Andrea Malagoli, University of Chicago (UC), is collaborating with Randy Hudson, ANL and University of Illinois at Chicago. UC's Robert Rosner leads the team studying such problems as convection in Sun-like and A stars, the evolution of magnetic fields in turbulent conducting fluids and the gravitational collapse of star-forming clouds.
The Caltech-JPL research exhibit (Booth R48) will be the site of meta-supercomputing demonstrations between those institutions and the San Diego Supercomputer Center. JPL's Advanced Laboratory for Parallel High Performance Applications (ALPHA) is developing the computations. The Research Exhibits will be accessible during the Gala Opening on Monday, December 4, 7 - 9 p.m.; on Tuesday, December 5 and Wednesday, December 6, 10 a.m. - 5:30 p.m.; and on Thursday, December 7, 10 a.m. - 4 p.m.
ALPHA staff also will be at the Cray Research exhibit (Booth 300) demonstrating interactive exploration of large Earth and space science image databases using JPL's CRAY T3D over the CASA gigabit-per-second testbed network. Synthetic aperture radar processing applications will be on display at the Intel exhibit (Booth 200). Industry Exhibits are open during the same times as the Research Exhibits.
ESS Project staff and investigators will be available for discussion at the National Coordination Office for High Performance Computing and Communications Program exhibit (Booth 640) on Thursday, December 7, 10 a.m. - 12 p.m. In addition, several visualization videos will be shown.
Co-investigators on another Grand Challenge team are analyzing the "Performance of a Parallel Global Atmospheric Chemical Tracer Model" during Applications: Ocean and Atmospheric Modeling, Thursday, December 7, 3:30 - 5 p.m. Sharon Smith and James Demmel of the University of California, Berkeley are constructing such a model for incorporation into the larger Earth System Model being developed under the direction of C. Roberto Mechoso of UCLA. At this stage, it primarily studies atmospheric pollution, including gas-phase and aqueous-phase chemistry, radiation, aerosol physics, advection, convection, deposition, visibility and emissions. The authors will describe the implementation and performance results on a variety of scalable parallel computers.
ESS Project staff members are presenting two papers at the conference:
"A Performance Evaluation of the Convex SPP-1000 Scalable Shared-memory Parallel Computer," Performance II, Wednesday, December 6, 3:30 - 5 p.m., will consider the findings of a set of empirical studies using both synthetic test codes and full applications for the Earth and space sciences to characterize this architecture's performance properties. The Convex Exemplar SPP-1000 is the first of a new generation of scalable shared memory multiprocessors incorporating full cache coherence, and this paper is the most in-depth study yet published on the system. The authors are Thomas Sterling, Phillip Merkey, Center of Excellence in Space Data and Information Sciences; Daniel Savarese, University of Maryland; Peter MacNeice, Hughes STX Corp.; Kevin Olson, Bruce Fryxell, George Mason University; and Clark Mobarry, GSFC.
JPL's John Lou will be speaking on "A Parallel Incompressible Flow Solver Package with a Parallel Multigrid Elliptic Kernel" at the Applications: Fluid Dynamics session, Thursday, December 7, 10:30 a.m. - noon. The objective of this work is to develop a parallel and scalable flow solver package for a variety of incompressible flow problems in physics and engineering applications, such as convective turbulence modeling in astrophysics, thermally driven flows in cooling systems and combustion process modeling. The kernel and flow solver scale very well to a large number of processors on the Intel Paragon and the CRAY T3D for computations with moderate granularity. Robert Ferraro, also of JPL, is co-author.
Paul Messina, Caltech and JPL, is a panelist in the HPCC Workshop "Federal Strategies and Programs in the Age of Information." The panel will convene Tuesday, December 5, 3:30 - 5 p.m. Earlier that afternoon, beginning at 1:30 p.m., federal agency accomplishments and plans for the High Performance Computing and Communications Program will be outlined, along with the National Science and Technology Council/Committee on Information and Communications' Strategic Implementation Plan. The Plan's six strategic focus areas will be discussed by the panel.
One of the difficulties in using scalable parallel systems for solving large-scale, real-world computational problems is a lack of useful system software tools. A Supercomputing '95 workshop "System Software and Tools for High Performance Computing Environments" will consist of an open review and discussion of the major findings of the "Second Pasadena Workshop" (January 1995) on this topic. That event focused on building an enabling infrastructure for migrating experimental results and tools from the research community to production-grade environments in support of the applications developers. The workshop is being held on Thursday, December 7, 8:30 a.m. - noon. Messina; James Pool, Caltech; and Sterling are the organizers.
Pool is chairing and Messina participating in the "Research Issues in Scalable I/O" workshop, Thursday, December 7, 1:30 - 5 p.m. Pool describes the multi-agency and multivendor Scalable I/O Initiative as "a coordinated attack on the biggest obstacle to effective use of teraFLOPS-scale computer systems by scientists and engineers: getting data into, out of and around such systems fast enough to avoid severe bottlenecks." The Initiative will determine applications' requirements and use them to guide the development of new programming language features, compiler techniques, system support services, file storage facilities and high performance networking software. The workshop will grant opportunity to identify research issues, evaluate progress and pose open problems.
A third ESS Project Grand Challenge team will have a poster on the "Implementation of an Atmospheric and Land Surface Model on Parallel Processing Supercomputers." It will be available for viewing in Poster Area 1, Number 04, on Wednesday, December 6 and Thursday, December 7, 10 - 11 a.m. GSFC's Max Suarez (principal investigator) and Daniel Schaffer and Carnegie Mellon University's Thomas Head have implemented parallel versions of the GSFC atmosphere and land surface models on the CRAY T3D, IBM SP-2 and Intel Paragon. The poster will include information on the development of these models and on their scalability on the different systems.
During the same time-slots in Poster Area 4, Number 42, the conclusions reached by "JNNIE: The Joint NSF-NASA Initiative on Evaluation" will be presented. To evaluate the performance and usability of various scalable parallel computers on representative applications, the National Science Foundation and NASA launched a two-year initiative in April 1993. Researchers from four NSF supercomputer centers and five NASA research centers collaborated to evaluate 22 applications on 19 types of computer systems. Sterling was one of the coordinators of JNNIE.
An electronic poster on "Estimation Algorithm for Satellite Altimetry Data on the CRAY T3D" will be on display in Poster Area 2, Number 21 on Tuesday, December 6 and Wednesday, December 7, 3 - 4 p.m. Parallel computers make possible extremely accurate and efficient ocean tidal models. This particular application uses TOPEX/Poseidon satellite data; it is based on differencing up to 80 equations from the data for each of 50,000 ground-track locations, forming the observation equations. Normal equations are then formed and solved simultaneously for 15,000 unknown parameters. The presenters are Andrea Hudson, Braulio Sanchez, GSFC; Dennis Morrow, Cray Research Inc.; and Nikos Pavlis, Hughes STX Corp.
ESS Grand Challenge Principal Investigator George Lake, University of Washington, is leading a session on "PetaFLOPS Computing" in Room 9 on Wednesday, December 7, 1:30 - 3:00 p.m. A petaFLOPS is equal to a million billion floating-point operations per second and is ten times the performance of all the networked computing capability in the United States. Individuals in government, academia and industry have realized that teraFLOPS systems will be inadequate to address many scientific and engineering problems that exist now, let alone applications that may arise in the future. As a result, the high performance computing community is examining the feasibility of achieving petaFLOPS computing over 20 years. Topics for this session include architecture and systems structure and applications and software. Messina and Sterling, co-authors with NASA Headquarters' Paul H. Smith of the MIT Press book Enabling Technologies for PetaFLOPS Computing, are participating.
Sterling is involved in two additional Birds-of-a-Feather (BOF) sessions:
"The Search for Diogenes' Lantern of MPP Benchmarking Truths" will be held in Room 8 on Tuesday, December 5, 6:30 - 8 p.m. A Supercomputing '94 panel on this subject attracted over 540 people. This BOF session will continue discussions of such issues as assessing MPP (massively parallel processing) performance, limitations of existing benchmark suites, user guidelines to assess MPP performance, performance trade-offs among modestly and massively parallel computers as well as high-end workstations, and deceptions in measuring MPP performance.
Having a 16-processor Convex Exemplar SPP-1 on site, GSFC is joining the Exemplar Alliance being organized by the University of Illinois' National Center for Supercomputing Applications. A BOF session is being held on Wednesday, December 7, 5:15 - 6:15 p.m. in Room 7B.
GSFC is a Level 0 Site on the conference's Information Wide Area Year (I-WAY), The center will be the entry point for ATDNet, the 2.5 billion-bit-per-second Advanced Technology Demonstration Network. A link through the NASA National Research and Education Network will provide access to the supercomputers and visualization facilities of six Washington, D.C.-area federal laboratories.
-end-
[Note to Reporters and Editors: During Supercomputing '95, Jarrett Cohen will be staying at the Radisson Hotel Harbor View, 619-239-6800.]