The Beowulf Cluster at the
Center for Computational Mathematics
Department of Mathematics
University of Colorado at Denver

Funded by the National Science Foundation

Getting Started

User Information

Beowulf Status Report

Beowulf Temperature and Load Report

Hardware

Software



More Pictures

Hardware

The cluster consists of 36 separate computers, called nodes,  and other hardware, all mounted in two 19" racks. The cluster has total of 72 CPUs, 72GB memory, and about 2TB of disks.

Master node: 2 Pentium-III processors 933MHz, 2GB memory, Tyan S2567 motherboard, mirrored 36GB 160 SCSI IBM Ultrastar disks, CD-ROM, floppy disk
35 compute nodes: 2 Pentium-III processors 933MHz, 2GB memory, Supermicro 370DER motherboard, 18GB 160 SCSI IBM Ultrastar disk, floppy disk
Mass storage: 1TB disk array (8 Seagate Barracuda 180GB disks 160 SCSI: striped and mirrored > 80MB/s sustained throughput), VXA Autopak-30 tape library.
Management interconnect: 100Mb/s switched Ethernet with 1Gb/s fiber uplink to master node
Compute interconnect: SCI Dolphin , 6x6 double torus, 220MB/s actual bandwidth, 5us latency
Power management: 4 APC Smart UPS 2200 with temperature sensors and network interface. Total power consumption about 5kW idle and 8kW fully loaded.
Console management: Serial console through 3 Perle IOLAN+ terminal servers
 

Software

Node operating system: Redhat 7.2, Linux kernel 2.4.9-34
Cluster management: Scali 3.0.1
Interconnect software (MPI implementation): Scali 3.0.1,  MPICH 1.2.3
Compilers&Debuggers: GNU (v2.96 and 3.1), PGI CDK (Portland Group v3.2 and 3.3), TotalView 5 (parallel debugger)
Job scheduling: ScaPBS - Scali packaging of OpenPBS 2.3 (Portable Batch System)
Development tools: PETSc (v2.1.1 and 2.1.3), Hypre (v1.6.0 and 1.7.6b), BlockSolve95, ScaLAPACK (v1.7)
Applications: Matlab 6.1 (release 12.1)
Coming soon: PGI CDK 4.0
 
 
Supplied by Western Scientific Marketing

The software was completely reinstalled as of March 1, 2002.
You may also use the  Mini Cluster (log onto oldmath.cudenver.edu).