*The Final Report from the ESS Round-1 Science Team was completed and posted on the World Wide Web (WWW).
*During November, the Round-2 Grand Challenge Investigator teams
of Peter Lyster/University of Maryland and David Curkendall/Jet
Propulsion Laboratory achieved their 10 Gigaflop/s performance milestone. As a result, Silicon Graphics/Cray Research also achieved their
10 Gigaflop/s performance milestone.
By March, all 9 teams had achieved 10 Gigaflop/s sustained performance on their code(s) as negotiated. Technical support by SGI/Cray
played a key role in many instances. The Tamas Gombosi/University of Michigan team and its milestone accomplishment was featured in HPCwire (3-7-97).
*In June, the Graham Carey/University of Texas at Austin team's code ran at 51.6 Gigaflop/s on Goddard Space Flight Center's (GSFC) 512- processor CRAY T3E even though the Cooperative Agreement allowed use of a computer twice that size. The John Gardner/Naval Research Laboratory (NRL) team achieved the first 50 Gigaflop/s performance milestone on their two codes.
*The first ESS Science Team meeting of Round-2 Investigators was held November 18 in Pittsburgh in conjunction with Supercomputing '96. A second Science Team meeting occurred April 7 in Atlanta.
*In December, ESS staff improved the ocean code performance for the Roberto Mechoso/UCLA team by 45 percent over the 1995 timing. The work involved replacing "if" statements with masking arrays, calling BLAS routines, and rewriting array syntax. Later, using one-dimensional data decomposition (in North-South direction) and one-pass load-balancing in physics, 24 percent performance improvement in the Atmospheric General Circulation Model (AGCM)/Physics module was achieved on 256 nodes of a CRAY T3D. Ten percent improvement was gained on the entire AGCM code.
*In February, the Gardner team submitted for the Software Exchange their documented 10 Gigaflop/s milestone code. This was the first submission of a 10 Gigaflop/s code. By mid-May, four teams had achieved their milestone for code submission.
*In March, the ESS WWW pages were updated to display the 117 Cooperative Agreement milestones for the Round-2 Investigators and SGI/Cray.
*GSFC Employee Development scheduled Directorate-funded CRAY T3E Training classes were held on-site April 28 through May 3 for the GSFC Labs.
*In August, ESS began working with GSFC Earth Sciences Directorate Laboratory Chiefs and NASA Headquarters Mission to Planet Earth and Office of Space Science program managers to formulate an allocation policy for the 20 percent of the CRAY T3E allotted for investigations drawn from the breadth of NASA science to prepare that community to use scalable parallel systems.
In January, GSFC released an announcement of opportunity to the NASA Earth and space science community and related space flight missions, requesting proposals by March 3.
Fourteen large allocations were decided and announced publicly April 21 for the four quarters beginning April '97, July '97, October '97, and January '98. A total of 106,250 CPU hours was allocated for the first 6-month period. Additional proposals will be selected via a similar proposal process every 6 months .
A set of the proposals receiving large allocations was provided to SGI/Cray staff to assist them in identification of the 10 codes that they will bring over to the CRAY T3E. The SGI/Cray milestone "Demonstrate 10 codes from the broader community on the T3E" will be met by assisting 10 of these investigations.
*In January, computations of convection in a hemispherical shell were performed on the CRAY T3D using the phase II SBIR parallel Nektonics Inc. code, which employs a spectral element formulation.
*In January, an MPI-based client/server communication structure was developed for a distributed comet-rendering application. The use of MPI-based client/server communication enables the computational work of the comet renderer to be parallelized in varying computing environments, including a network of heterogeneous workstations and a network of workstations and massively-parallel computers.
*In the winter, members of the ESS/GSFC inhouse team began developing an Adaptive Mesh Refinement (AMR) package to support two- and three-dimensional models that use logically Cartesian meshes. The package covers the computational domain with uniform mesh sub-blocks, with each sub-block refined to meet the applications resolution requirements. The sub-blocks belong to a tree data-structure. The package manages the construction and modification of this tree and all communications required between blocks and processors.
*ESS support for the MIDAS project ended at the close of FY96. MIDAS is now being funded by a variety of other sources.
*In October, SGI/Cray installed a 512-processor CRAY T3D system at GSFC as part of a Cooperative Agreement (CAN). The system was integrated into NASA Center for Computational Sciences (NCCS) user accounting, networking, and operations. CAN Investigator teams received user accounts and began using the T3D, named "seymour" (in memory of Seymour Cray). A Sun server for Cray on-line documentation was also installed. A full set of T3D WWW Pages was developed and brought on-line for the HPCC CAN T3D user community. It was accessible via links from both the NCCS and the HPCC ESS Home Pages. This system was listed as #30 on the Netlib's November 1996 TOP500 list of supercomputers in the world.
In December, SGI/Cray's new CAN Applications Support staff members began work: Dr. Tom Clune was formerly a member of the University of Colorado group of Andrea Malagoli's/University of Chicago Investigator team, and Mr. Spencer Swift is completing his Ph.D. dissertation work with Investigator Graham Carey/University of Texas.
SGI/Cray met their December milestone by providing CRAY T3E training for Investigator teams and NASA HPCC staff and identifying Investigator codes that will achieve 50 Gigaflop/s.
In June, SGI/Cray completed installation of a 512-processor CRAY T3E system at GSFC. This system has 64 gigabytes of memory, 80 gigabytes of disk, and a peak performance of 306 Gigaflop/s. Based on the June 1997 TOP500 list that uses the LINPACK benchmark, this T3E ties for second in the U.S. and sixth in the world. A formal dedication ceremony was held May 14 to name the T3E "jsimpson" in honor of Joanne Simpson, GSFC's chief scientist for meteorology.
*In January, ESS placed an order with Cray/SGI via the SEWP II contract for a Powderhorn tape silo with eight Timberline and four Redwood drives to be directly attached to the T3E and operate under DMF.
*In January, a demo version of The Portland Group's HPF compiler for the T3D/T3E was installed on the T3D at GSFC.
*In December, the Godiva (GODdard Instrumentation Visualizer and Analyzer) software system, developed by Terry Pratt/CESDIS, was running on the T3D testbed, Beowulf parallel workstation systems, and Sun workstations. Godiva provides a flexible software base for analyzing various performance properties of large science codes.
In March, results from evaluation studies of two codes (FCT and CRUNCH-3D) from the Gardner team were reported back to that team, including a suggestion for an improved parallel transpose algorithm for CRUNCH-3D.
*In January, ESS powered down the 16 processor Hewlett-Packard/Convex Exemplar SPP1000 at GSFC. Its two generation old technology was no longer supported in software and could not be upgraded.
*In March, the National Science Foundation (NSF) announced that the proposal team led by the San Diego Supercomputer Center and involving Caltech had won one of the two NSF supercomputer center recompete awards under the Partnerships for Advanced Computing Infrastructure (PACI) program. This completes the funding triad that brought the largest HP/Convex Exemplar SPP2000 (256 processors) to Caltech this summer and later will bring the first SPP3000 delivered to a customer site. The sustainable performance of the SPP2000 is expected to be about four times that of a Cray T3D on a per processor basis. This partnership among NSF, NASA (ESS and Office of Space Science), and the Jet Propulsion Laboratory (JPL) will afford the NASA HPCC community access to one of the largest current generation cache coherent, shared distributed memory machines to explore the viability of that programming model.
*In June, GSFC assembled the Phase I skeletal system of the Beowulf Scalable Mass Storage Research project (T-Racks), which is partially funded by Defense Advanced Research Projects Agency (DARPA). The system includes 60 Pentium Pro processors, 3.7 gigabytes of RAM, and 280 gigabytes of disk, interconnected with 100 megabits/second (Fast) Ethernet. The goal of is to assemble a terabyte mass storage server out of commercial commodity-off-the-shelf microprocessors and achieve a gigabyte/second of aggregate data bandwidth.
*By October, Los Alamos National Laboratory and Caltech/JPL had assembled Beowulf systems linking 16 Pentium Pro processors. Mike Warren/Los Alamos assembled a Beowulf system and ran the Salmon/Warren tree code cosmology simulation on 10 million particles. He achieved a sustained performance of around 1.1 Gigaflop/s for a cost of under $60,000. A 10 million particle tree code developed by John Salmon/Caltech was ported to the Caltech/JPL Beowulf and achieved sustained performance of 1.26 Gigaflop/s for a cost of approximately $50,000.
In November, Linux kernel support for Global Process ID Space was ready for mainstream distribution, with inclusion in the stock kernel.
In November, ESS submitted Beowulf as a potential nominee to Discover Magazine's Computer Innovation Awards, and the Awards organizer decided to nominate it.
Beowulf systems from Los Alamos and Caltech, brought to Supercomputing '96 (November), were joined together into a 32-processor Beowulf (worth around $100,000) on the exhibit floor and ran Warren/Salmon tree code problems at around 2.2 Gigaflop/s.
The December 13 issue of Science carried an article entitled "Do-It-Yourself Supercomputers" that presents the Beowulf project and its recent performance price breakthrough.
In a joint effort with the Caltech Beowulf team, a "How to Build a Beowulf" tutorial was developed. The 4-hour tutorial was given at the Cluster Computing Conference (CCC97) held in March at Emory University in Atlanta. This series of lectures included demonstration of a complete eight-node Beowulf Pile of PC's running an N-body simulation. Hardware and software assembly, installation, and use were discussed, with numerous demonstrations. As a result of the tutorial's success, it is planned to give a similar tutorial at Caltech and again at Supercomputing '97 in San Jose.
Dan Ridge and Don Becker/CESDIS worked with RedHat Software to produce multi-platform CD's of Beowulf Linux. This year's CD includes Global Process ID Space system software, which brings tools for parallel computing to the mainstream for the first time. RedHat is shipping 300,000 CD versions of this software monthly. Other Parallel Linux enhancements are forthcoming as RedHat packages. NASA and CESDIS are fully identified in the CD, which is available at Comp USA and other retail stores.
In June, Beowulf Gigabit Ethernet was functional. Becker wrote the device driver software for this connection that can enable data transfer up to 10 times faster than Fast Ethernet. The software is available on the Beowulf WWW site. A beta test site agreement was negotiated with Packet Engines Inc., who produce Gigabit Ethernet hardware. GSFC acquired six of their cards and serial #1 of their Gigabit Ethernet repeater to build an eight node sub-cluster and determine whether or not this technology will satisfy high bandwidth requirements.
*In December, Robert Ferraro/JPL chaired the second National HPCC Software Exchange (NHSE) Technical Committee Meeting at Rice University. Representatives from NASA, the Department of Energy, the National Institute of Standards and Technology, NSF, DARPA, and the Department of Defense participated on the committee and found that NHSE has made substantial progress since the last Technical Committee meeting. A report from the Technical Committee was produced and circulated to the sponsoring Agencies.
In November, the PTLIB (Parallel Tools Library), HPC-Netlib (a high-performance branch of the Netlib software), and Chemistry Software and Information Services (CSIR) repositories came on-line. They are accessible from the NHSE home page.
The NHSE's Repository in a Box Maintainer's Toolkit and the NHSE Repository Planning Guide became available via the NHSE home page.
*The language definition and compiler implementation phase of the aCe (architecture independent C) programming paradigm was completed, and the application development and performance evaluation phase began.
*In October, a distributed visualization prototype system delivering multiple color frames (2-3, 512 by 512 color frames) per second was demonstrated over the Internet from the HPCC site at Minnesota to an end-user at JPL.
In November, the ParVox API achieved super-linear speedup with increasing input problem size and a fixed output image size. It also achieved near linear speedup with both increasing problem size and increasing image size.
The Acoustetron II sound generator, now operational in the Virtual Environments Laboratory at GSFC, allows a user to sonically probe data while wearing the CyberGlove. The values at the fingertip location are mapped to varying sound characteristics (e.g., pitch), which can appear to emanate from the fingertip.
*In March, OC-12 (622 megabits/second) connectivity existed among NRL, the Defense Intelligence Agency, the National Security Agency, and GSFC on the Advanced Technology Demonstration Network (ATDNet). GSFC has further interconnected that OC-12 system with the Advanced Communications Technology Satellite (ACTS) High Data Rate Terminal (HDRT) at GSFC and is conducting/supporting experiments as part of the "622 Megabits/Second Network Tests Between ATDNet and MAGIC Via ACTS," ACTS Experiment #118, a.k.a. AAMnet.
*In March, the Office of Space Science's Ramon DePaula and LeRC's Mike Zernic announced that they were providing GSFC with augmentations of FY97 funds to support efforts in the AAMnet ("622 Megabits/Second Network Tests Between ATDNet and MAGIC Via ACTS" ACTS Experiment #118).
*On October 15, the LRIS instrument at the Keck telescope (Mauna Kea, Hawaii) was interactively operated from a workstation located at the Caltech campus in California. The communications link to Hawaii consisted of a combination of satellite (ACTS), fiber optic (SONET OC-3), and microwave (DS-3) link technologies. In the demonstration, LRIS test images were transferred at an average rate of 15 megabits/second to a Sun SPARC 20 workstation located at Caltech.
*In November, staff enabled NASA's first 622 megabit/second ATM wide area network use providing OC-12c links among the ATDNet, the ACTS's HDRT, and the first OC-12c attached computer at GSFC.
*In January, staff enabled the National Library of Medicine (NLM) as the first non-GSFC ATDNet site to begin tests over ACTS via a virtual link established across ATDNet between GSFC and NLM.
*In January, GSFC was invited by DARPA to become a node in the DARPA-initiated, n x 10 gigabit/second, wave division multiplexed (WDM) based MONET testbed that will be integrated into the ATDNet's infrastructure.
*In February, in collaboration with NREN Project Manager Christine Falsetti/Ames Research Center (ARC), GSFC sent a request to the ESS Investigators explaining NREN and its potential interconnection to the vBNS (NSF backbone network) and soliciting abstracts for applications requiring NREN connectivity. Several Investigators provided abstracts.
*In February, ACTS operations staff solved a recurring loss of synchronization by providing a third HDRT on-line functioning as the master clock synchronizer, which then enabled HDRT's at GSFC and Kansas City to sync up at 622 megabits/second. In June, the first-ever OC-12c end-to-end benchmark performance tests between Kansas University and GSFC achieved over 120 megabits/second.
*In March, GSFC achieved a new network throughput "high water mark" of 310 megabits/second over ATM using the 622 megabit/second Network Interface Card (NIC) in a Sun UltraSPARC.
*In March, staff achieved a new ACTS "high water mark" in end-to-end data transfer performance over OC-12 (622 megabits/second) of over 215 megabits/second between UltraSPARC's at JPL and LeRC.
*In July, GSFC hosted the Seventh Annual NASA Summer School for High Performance Computational Physics.
*Six NASA Graduate Student Research Program (GSRP) Fellows and three National Research Council (NRC) postdocs were supported by ESS. Five GSRP Fellows completed training under ESS support from prior years.
ESS/JPL NRC Postdoc Charles Norton was appointed to the U.S. delegation for upcoming FORTRAN 2000 standardization meetings in Las Vegas due to his research work on OO programming using FORTRAN 90.
*In October, the final report (version 1.0) for the combined workshops on Petaflop/s Architecture (PAWS) and Petaflop/s System Software was released. This major work comprises over 500 pages involving technical contributions of close to a hundred experts in a broad range of fields. This is the foremost current work in the emerging interdisciplinary science of Petaflop/s.
*In October, ESS sponsored Frontiers '96, the Sixth Symposium on the Frontiers of Massively Parallel Computation, held in Annapolis, MD. Noteworthy were presentations by all eight NSF Petaflop/s point design study teams and a ceremony officially handing the Massively Parallel Processor (MPP) over to the Smithsonian.
Staff organized and carried out the second Petaflop/s Frontier workshop at Frontiers. It is the only forum directly related to Petaflop/s computing that is open to the general community; all other past organized activities being by invitation only.
Chip Weems/University of Massachusetts at Amherst will serve as Chair and Bill Carlson/Center for Computer Science will be Vice Chair of the Frontiers '98 Symposium.
*In June, ESS was presented at the fourth Independent Annual Review (IAR) of HPCC held at ARC. The May 1997 Level Two ESS Project Plan was distributed to the Panel.
*ESS participated extensively in Supercomputing '96 (November 17-22, Pittsburgh), including taking part in the first NASA HPCC research exhibit. Two news releases were distributed to the national media.
*The annual ESS Center Management Meeting between GSFC and JPL was held during Supercomputing '96.
*In January and February, ESS management met with representatives of the National Oceanic and Atmospheric Administration (NOAA)'s National Centers for Environmental Prediction to discuss using the ESS CRAY T3E to test code performance and scalability in preparation for their production use of a class-8 supercomputer (being procured in FY98 and installed at GSFC in Building 28). NOAA representatives included Eugenia Kalnay, Chief Scientist for the National Weather Service. It was proposed that NOAA-funded scientists be treated as the 10th Grand Challenge Team.
*In February, GSFC Center Director Joe Rothenberg approved the NASA HPCC Level-1 Program Plan developed by ARC.
James R. Fischer
Goddard Space Flight Center
James.R.Fischer.1@gsfc.nasa.gov
301-286-3465
Robert Ferraro
Jet Propulsion Laboratory
ferraro@zion.jpl.nasa.gov
818-354-1340
Table of Contents | Next Section -- Applications