HIGH PERFORMANCE COMPUTING CENTER (HPCC)

The High Performance Computing Center evolved from the former Physics Detector and Simulation Facility, later the Parallel Distributed Systems Facility (PDSF).

The workstations in the ranch (farm) normally run in coarse-grain parallelism and are available for batch and interactive use. Batch queuing is provided using Network Queuing System (NQS). Applications that require fine-grain parallelism can run under Parallel Virtual Machine (PVM). The large amount of disk space allows use of advanced distributed databases and information retrieval systems. Due to its modular architecture, the system is easily upgradable and expandable. It can be used as a test bed for emerging hardware and software technologies and new interoperability concepts.

The Parallel Distributed Systems Facility (PDSF) provides 12,000 MIPS computing power on a UNIX workstation ranch (farm). A total of 124 processors are organized in groups that are connected by Fiber Distributed Data Interface (FDDI) and Ethernet. Several types of stations are used, including HP 9000/735, Sun SPARC10, and SGI Challenge L and SGI 4D/360. Each workstation has at least 64 MB of RAM and 1 GB of disk storage. Three SGI Challenge L file servers provide an additional 160 GB of disk space. Two Summus STL-2300-12 tape robots are used for archiving.

The Intel iPSC/860 Hypercube provides 2112 MIPS of computing power, using 64 tightly coupled nodes, each with 8 MB of RAM. Total storage space is 5 GB. The Hypercube can be used for massively parallel scientific and technical applications in image processing, ray tracing, climatology, and hydrodynamics.

HPCC Configuration and Specifications

Compute Systems Configuration (a 28480 bytes Gif file)

Networking Configuration (a 31737 bytes GIF file)

Compute Systems Specifications (a 158562 bytes GIF file)

PDSF History of Growth

1991: PDSF operation begins at 1000 MIPS of parallel processing power and 40 GB

of disk space. Hypercube operation begins with 64 nodes and 2112 MIPS of processing power.

1992: PDSF is increased to 2000 MIPS of parallel processing power; disk space

increases to 80 GB.

1993: PDSF is increased to 8000 MIPS of processing power with 240 GB of disk

on-line.

1994: PDSF is increased to 12,000 MIPS of processing power. Potential growth:

16,000 to 20,000 MIPS of parallel processing power.

Features

High-performance RISC workstation Compute Servers and File Servers.

14,000 MIPS of computing power.

More than 200 GB of on-line disk storage.

500 GB tape robotic system.

Operating System support.

Network Queuing System (NQS) for batch execution.

Multiple UNIX platforms for major vendors including Sun, HP and SGI.

Access to Compute Servers and File Servers via both Ethernet and FDDI networks.

Acess to X-Windows interactive sessions.

Access to the Internet via dedicated T1 circuits.

SSC Scientific and Technical Electronic Repository