PSC News Center

Download the PDF version of this article as it appeared in Projects in Scientific Computing, 2003.


Two new systems established residence in the PSC machine room this year. Processor for processor, these systems — Hewlett-Packard GS1280 AlphaServers — are even more powerful than LeMieux, PSC’s workhorse 3,000 processor terascale system, which delivers much of the public-domain computational research in the country.

During development, HP referred to the GS1280 as the Marvel, and PSC’s twin Marvels are among the first to roll out of HP production. Both house 128 of the newest generation of the powerful Alpha processor, the EV7. Both provide 512 gigabytes of shared memory — each soon to be configured as two 256-gigabyte, 64-processor units.

GS1280s at PSC
Rachel CarsonJonas Salk

What's in a Name?: Meet Jonas & Rachel
Since Mario, the CRAY C90, from the early 1990s, PSC tradition has been to name its systems, and the newly arrived Marvels presented an opportunity. This time PSC turned from sports to science. The twin Marvels are Rachel and Jonas, for Rachel Carson, the southwest Pennsylvania native who jump-started modern environmentalism with her 1962 book Silent Spring, and Jonas Salk, whose polio vaccine, developed at the University of Pittsburgh Medical School, in 1955 halted the spread of this crippling disease.

The only difference between these capable Marvel twins is that one of them, dubbed Jonas, is dedicated to NIH biomedical research, while the other, Rachel, supports NSF science and engineering. Together, they complement LeMieux with new capabilities for scientists and engineers nationwide.

Along with a “shared memory” architecture, both new systems have exceptional “memory bandwidth” — the speed at which data transfers between hardware memory and the processor. Benchmark tests have demonstrated that the GS1280 memory bandwidth is five to ten times greater than comparable systems.

“Such a shared memory, high-performance system is not now available to biomedical researchers,” said Marjorie A. Tingle, director of the Shared Instrumentation Grant Program at NIH’s National Center for Research Resources, which awarded $1.3 million to PSC for Jonas. “This new resource will enable them to study particularly demanding data-intensive, memory-intensive and compute-intensive problems that are currently beyond reach.”

Funding for Rachel is part of a $35 million NSF grant to PSC and its two sister supercomputing centers in October 2002. This grant supports continued development of the TeraGrid (see facing page) and provides for harnessing LeMieux and Rachel into this evolving national grid.

“The memory structure of this system represents a significant advance for scientific computing,” said PSC scientific directors Michael Levine and Ralph Roskies in a joint statement. “It will make a difference in several important areas, including quantum chemistry and genomics.”

PSC Connects to the TeraGrid


The nation’s most powerful, academically-based computing resource — LeMieux — is now linked with supercomputers in Illinois and California as part of the TeraGrid, an integrated national system of cyberinfrastructure. PSC network engineers in August implemented and tested a light pipeline that connects PSC to the Chicago hub of the TeraGrid’s high-speed “backplane,” which joins Chicago and Los Angeles.

Tereagrid illustration.

Funded by NSF, the TeraGrid is a multi-year effort to deploy the world’s fastest, most comprehensive distributed-computing infrastructure for open scientific research. Analogous to an electrical power grid, the TeraGrid will make computational power available to scientists and engineers nationwide as a seamless resource, without regard to physical location of computing systems.

Application Gateways
PSC network engineers have developed a unique and effective approach to link the TeraGrid backplane and LeMieux. The fiber-optic line — called a lambda — connects directly to a router, supplied by Cisco Systems, in the PSC machine room. From there, however, rather than configuring all 750 of LeMieux’s AlphaServers with a direct connection, a much smaller number of machines (from 20 to 30) will bridge from the router to LeMieux. Called an “Application Gateway,” this interface will run software, developed by PSC, that transfers data between the TeraGrid lambdas and LeMieux’s internal network.

Implemented in February 2003, the TeraGrid’s Chicago-Los Angeles backplane moves data at 40 gigabits per second — 600,000 times faster than typical Internet dial-up — the fastest network in the world. It is referred to as a backplane — a circuit board with plug-in slots for other devices — because the TeraGrid is conceived of as a single national machine for science and engineering research, with powerful systems in different places that plug-in to the backplane. The initial Pittsburgh to Chicago link of 10 gigabits per second, will triple to 30 Gbps within the next year.

LeMieux, different in both processor and operating system than existing TeraGrid systems, poses the challenge of “interoperability” — to create a grid environment integrating heterogeneous systems. The TeraGrid has established a PSC-led Interoperability Working Group to develop the software needed to meet this challenge.

“Many of us in many places are working together to create the TeraGrid,” said Rick Stevens of Argonne National Laboratory and the University of Chicago, TeraGrid project director. “This work will empower U.S. research in science and engineering for years to come. The Pittsburgh to Chicago connection is a major step toward bringing this vision of integrated national cyberinfrastructure into reality.”

More information:


PSC News Center
All of PSC

Jim Kasdorf

Jim Kasdorf, PSC director of special projects