CSC Getting Started Guide

CSC Computational Resources

The CSC operates several clusters and standalone systems for research computing and software development tasks. The two primary computational clusters are named Hemisphere and Occam. We make every effort to keep the systems continuously available, although we occasionally take them offline for maintenance or reserve capacity for special projects. Resources on the clusters are shared using a batch scheduler called PBS. In addition to the clusters, CSC runs a 64-bit Xeon system called Toaster. As an experimental system, Toaster is available for use by request only.

Hemisphere Cluster

The 64-node Hemisphere cluster is designed for debugging parallel jobs using MPI, running small parallel programs, and running large-quantity independent simulations such as parameter studies. Because of the number of nodes on Hemisphere, some users like to use the cluster to run a large number of single-processor jobs simultaneously. This is perfectly acceptable, but the queuing system has been configured to throttle jobs to prevent a single user from hogging the system with hundreds of single-processor jobs. Single-processor jobs may also be run on Occam.

Hemisphere supports a wide range of compilers including gcc 3.3, the Intel 9.x suite, and the Portland Group PGI 5.2 suite. With this collection of compilers, we have been able to compile almost everything on Hemisphere. Information about using the compilers is available on the Hemisphere web page.

System:	Hemisphere Cluster
Front End:	hemisphere.cs.colorado.edu
Nodes:	64
Processors:	2 Intel Xeon 2.4 GHz per node
Memory:	2 GB per node
Interconnect:	Gigabit Ethernet – MPICH

Occam Cluster

The 27-node Occam cluster is designed for mixed-use computing including parallel programs and single-processor job batch processing. Occam is currently less utilized than Hemisphere, so job turnaround time is usually less on Occam.

Occam is networked with gigabit Ethernet and provides MPICH for run parallel jobs. In our experience, the Occam Ethernet is higher latency and lower throughput than the Ethernet on Hemisphere, so tightly coupled jobs generally perform better on Hemisphere. (Of course, for a production parallel code, consider requesting an allocation on NCAR’s Frost supercomputer.)

Occam has both gcc 3.3 and the IBM XLF/XLC compiler suite installed. As a PowerPC system, Occam supports both 32-bit and 64-bit executables. Compiling using gcc without any flags creates inefficient 32-bit code. You must set compiler flags to compile in 64-bit mode: see the Occam web page for details. In general, the IBM XLF/XLC compiler suite performs the best optimizations for the IBM PowerPC 970 processors in Occam.

System:	Occam Cluster
Front End:	occam.cs.colorado.edu
Nodes:	27
Processors:	2 IBM PowerPC 970
Memory:	2.5 GB per node
Interconnect:	Gigabit Ethernet – MPICH

Toaster Experimental System

Toaster is a single node Intel Xeon EMT64 system with 8 GB of RAM. This system is capable of running jobs that do not fit in the memory available on other systems. Toaster is an experimental system and is reserved for interactive use by request only.

Because Toaster is an Intel EMT64 system, it supports both 32-bit and 64-bit code. Both the gcc 3.3 compiler and the Intel ICC/IFORT 8.x compiler suite are available. When compiling, make sure to check the compile flag documentation and generate code optimized for this architecture. See the Toaster web page for compilation instructions.

System:	Toaster SMP
Front End:	Use hemisphere or occam to connect, then ssh to “loaner”
Processors:	2 Intel Xeon EMT64 3.2 GHz
Memory:	8 GB

‹ prev | 1 2 3 4 5 6 7 8 | next ›