Science and Visualization

Computing Remotely at the Supercomputing Centers

Frances Bauer

Researchers requiring large amounts of memory or computing time, with problems that cannot be adequately addressed at the ACF facilities, can obtain allocations of time at either of two National Science Foundation (NSF) Supercomputing Centers.

NCSA, the National Computational Science Alliance at the University of Illinois (www.ncsa.uiuc.edu), and NPACI, the National Partnership for Advanced Computational Infrastructure (www.npaci.edu) at the San Diego Scientific Computing Center (SDSC), are the two centers established in 1997 to continue the NSF support for supercomputing. Detailed information about the centers can be obtained by accessing their websites.

NCSA Resources (Available at the University of Illinois)

An HP/Convex Exemplar SPP-2000(X-Class), based on the HP PA-RISC 8000 Processor.

A Silicon Graphics CRAY Origin2000, based on the MIPS R10000 processor. The SGI CRAY Origin2000 has 512 processors with 128 gigabytes of memory. Each processor is 195 MHz. Peak performance per processor is 390 MFLOPS. The operating system is IRIX 6.4. Load-sharing software is available for batch, and native MPI and PVM are available for message passing.

A Silicon Graphics Power Challenge Array, based on MIPS 8000 and MIPS R10000 processors. This networked array of bus-based shared-memory supercomputers consists of an eight-processor R4400 machine with 2 gigabytes of memory, which is the file server; 68 MIPS R8000 processors, each 90 MHz with 16 gigabytes of memory and 111 gigabytes of scratch disk space; and 72 MIPS R10000 processors, each 195 MHz with 18 gigabytes of memory. The operating system is IRIX 6.2 , the batch scheduling software is LSF, and MPICH and PVM are the software used for message passing.

Allocations at NCSA

There are three committees to which an eligible Principal Investigator can submit a proposal: the Small Allocations Committee (SAC), the Alliance Allocations Board (AAB), or the National Resource Allocations Board (NRAB).

The SAC grants time allocations of from 1,000 to 4,000 Service Units (SU) per year. (An SU equals the product of the normalizing factor, the service level and CPU time. The normalizing factor is 1.2 for the MIPS R8000 and 1.3 for the R10000. The service level equals 1.0 for a standard batch and 2.0 for a high-priority batch. The service level for interactive computing is 1.5.)

The committee allocates NCSA resources only. Inexperienced researchers can apply for small amounts of time which can be used for exploratory purposes.

The AAB grants one-year allocations of 4,000 to 40,000 SUs at one of the following locations: NCSA, Boston University, Maui High Performance Computing Center, the University of Kentucky, and the University of Wisconsin at Madison.

The NRAB grants one-year allocations of over 40,000 SUs on computers available at the Alliance Institution and at the NPACI Institutions. Faculty members who want to use high-performance computers in their classes may apply for 1000 SUs per course.

Researchers can no longer obtain startup accounts since the demand for computing time is so great. However, as soon as a completed application is received and before peer review, 10 percent of the requested time is allocated so that the researcher can get started. You may apply to both centers for allocations for the same project, but it should be noted that you have done so on your application.

Available NPACI Resources

Any U.S. researcher or educator is eligible to apply for NPACI resources. High school, undergraduate or graduate students may obtain allocations by having their advisors apply.

The Parallel Vector Plat-form (PVP) machines available are a Cray T90, located at SDSC, with 14 processors and a total of 4 gigabytes of memory; a Cray J916 at the University of Texas with 16 processors and 4 gigabytes of memory; and a Cray J95, also at the University of Texas, with 1 gigabyte of memory.

The Scalable Parallel Platform (SPP) machines available are a Cray T3E at SDSC with 256 processors and 128 MB per processor, a Cray T3E at the University of Texas with 64 128-MB procesors, and an IBM SP at SDSC with 128 MHz and 256 MB per processor. The IBM SP floating point operations are faster than the Cray T3E.

There is an HP/Convex Exemplar available at Caltech with 256 nodes. The University of California at Berkeley has a cluster called NOW, which is a network of workstations.

Allocations at NPACI

Moderate allocations of resources are made by the NPACI Allocation Committee on a quarterly basis. Proposals must be submitted 60 days before the quarterly meeting.

Very large allocations are made by the NRAC on a semi-annual basis. Expedited allocations for small amounts of time can be requested at any time and will receive expedited review. An expedited allotment for the IBM SP is 2,000 process hours. A large allocation is 20,000 process hours. Larger amounts of time must be requested from the Metacenter Allocation Committee, which considers requests for large computational resources across multiple centers. [ C ]

Frances Bauer is a research scientist in ACF's Scientific Computing and Visualization group.
frances.bauer@nyu.edu

Posted May 18,1998