News

The San Diego Supercomputer Center (SDSC) has embarked on a major research collaboration with Cray Research, Inc. (CRI), to broaden the applicability of Cray's scalable parallel systems, the current CRAY T3D system and the forthcoming system (code named "T3E") to be officially announced by Cray tomorrow. In conjunction with this collaboration, SDSC recently acquired a CRAY T3D system and doubled the memory of its existing CRAY C90 vector supercomputer. SDSC has ordered a T3E system that is expected to be installed next year.

Scalable parallel processing harnesses hundreds and thousands of processors in parallel, applying them to a single problem. Many application codes today are not parallelized and so cannot take advantage of the power of parallel computers.

Efforts already underway at SDSC include parallelization and optimization of widely used applications in quantum chemistry and scientific visualization, such as GAMESS and Renderman. These efforts will expand to include applications in biomedicine as well, SDSC officials said.

CRI is expected to announce leading-edge scalable input/output (I/O) and networking technology as part of its announcement tomorrow of the T3E system. Opportunities to use the CRAY T3D and T3E systems for "data mining" of terabyte-sized data sets will also be investigated in SDSC's new Massive Data Analysis Testbed. This multi-vendor, heterogeneous computing environment will be interconnected by Cray's new I/O and networking technology, and Cray and SDSC will work together to demonstrate the effectiveness of this technology in such an environment for data-mining applications. Data mining refers to looking for correlations among data sets, thereby converting data into knowledge. In SDSC's case, these data sets are hugein the multiple terabyte-size range. Data mining is an important application area to improve decision making in government and private industry including the retail sector, banking, insurance, and telecommunications. "This agreement marks a watershed in SDSC's long-standing relationship with Cray Research," says SDSC director Sid Karin. "It points to much closer collaboration between the two organizations in the future in support of our nation's research community."

According to Robert H. Ewald, Cray Research president and chief operating officer, "We are pleased to see that SDSC has acquired our scalable parallel systems and to collaborate with SDSC on these important areas to parallelize critical applications and bring new capabilities to the research community and industry. We are particularly excited about demonstrating the usefulness of Cray's leading-edge supercomputing technology for data-mining applications. The complexity of these data problems, most commonly dealt with by commercial companies, have evolved to the point where they require the powerful capabilities of supercomputer technologies."

SDSC's 128-processor CRAY T3D system was installed last month. It has an aggregate peak speed of more than 19 gigaflops (billion floating-point operations per second) and 64 million bytes of memory for each processor, a total of 8 gigabytesfour times the total memory of SDSC's just upgraded CRAY C90 system. The CRAY T3D also has 63 gigabytes of disk storage.

"We're excited about the CRAY T3D system," says Wayne Pfeiffer, SDSC deputy director for research. "It significantly increases the computing power and memory available to our research community and represents another step in our move to scalable parallel computing." The relatively large memory will be particularly useful for chemistry applications, many of which have data storage requirements that scale with the square of the size of the problem.

The T3D features the industry's fastest interconnect technology, with high bandwidth and very low latency. Measurements by SDSC staff have shown a node-to-node bandwidth of 140 MBps (for large data transfers) and a latency of 1 s (for small data transfers). Such high communication performance is especially important for applications that exchange large amounts of data between nodes, such as the general circulation models used in climate modeling.

Bob Leary, an applied mathematician at SDSC, ran the first production calculation on the T3D to minimize the energy of a cluster of nearly 50,000 particles interacting via a Lennard-Jones potential. This project represents an excellent example of modeling phase transitions and structural changes in weakly interacting systems. With 385 iterations each taking 94 seconds on all 128 processors, the calculation required 10 hours at a sustained computational rate of about 2 Gflops. The calculation produced coordinates of a minimum in the potential energy surface corresponding to an icosahedral (20-sided) configuration.

Other research projects that have begun using the T3D include general circulation modeling at the Scripps Institution of Oceanography, quantum Monte Carlo systems studies at UC Santa Barbara, and population genetics studies at the University of Southern California.

SDSC, a national laboratory for computational science and engineering, is sponsored by NSF, other agencies, the State and University of California, and private organizations; is affiliated with the University of California, San Diego; and is administered by General Atomics. For more information, see http://www.sdsc.edu or contact Ann Redelfs, SDSC, redelfs@sdsc.edu, 619-534-5032.

For more information, contact:

Ann Redelfs
SDSC
619-534-5032
619-534-5113 (FAX)
redelfs@sdsc.edu

Mardi Larson
Cray Research, Inc.
612-683-3538
schmma@garnet.cray.com