Numerically Intensive Computing
Academic and Research Computing (ARC) provides two facilities
that are dedicated to numerically-intensive research computing:
You can apply for access to both services by completing a
Batch Services Access Request Form
and turning it in to the VCC Help
Desk. Applicants who are not faculty members must have a faculty
member sign the form. You will be notified by e-mail when access
is approved, and you must then wait 24 hours (necessary for
password propagation) before attempting to use either service.
Both services can be used for brief interactive sessions for
program development and debugging, but production runs should be
made by submitting batch jobs using the Distributed Queueing
System, DQS. You need to submit your job to the appropriate queue
based on your requirements for memory, CPU time, and
software.
You can use the Unix qstatus command to see what queues are
available.
Quick Study #30: Batch Processing for NIC Users
describes DQS in more detail, and
contains examples showing how to use it. Both services participate
in the campus AFS file system, and you access them using your RCS
username and password.
For help in using the batch services, send email to
nic-support-l@lists.rpi.edu.
More information about numerically-intensive computing at
Rensselaer and in general can be found at the following web sites.
The serial queues
The serial queues are intended primarily for long-running research
calculations that do not use parallel processing. The serial
queues run on the following hardware:
-
3 IBM 43P Model 260 machines each with two 200 MHz Power3
processors
-
2 IBM 7026-B80 machines each with two 375 MHz Power3-II processors
These machines run AIX 4.3, and support DQS queues with RAM limits
of 256 MB or 512 MB and temporary disk space limits of 508 MB or
1024 MB (use the qstatus command to find out more details about the
serial queues).
The landing pad for interactive use of this service is
camaro.nic.rpi.edu
The parallel cluster is intended primarily for parallel research
calculations, with or without message passing. Message passing is
provided by MPICH, a public-domain version of the Message Passing
Interface library, over 100 Mb/sec ethernet connections between the
machines. The parallel cluster consists of 8 PCs each with two 863
MHz Pentium-3 processors and approximately 1.7 GB of local
temporary disk space. These machines run Red Hat Linux 7.0 and
support DQS queues with RAM limits of 256 MB (use the qstatus
command to find out more details about the parallel cluster
queues).
In addition to the local temporary disk space, the cluster has a
shared scratch pool in /colony/scratch1 containing approximately
17 GB.
The landing pad for interactive use of the parallel cluster is
miata.nic.rpi.edu
Because the batch services run on RCS machines, most RCS software
is available on them. However, some RCS software that is unrelated
to numerically intensive computing might not be available, and some
software might be available on one service but not the other.
Here is a table listing some of the available software of interest
to research users.
| serial queues | parallel cluster |
| NAG Library Mark 19 for AIX
|
NAG Library Mark 19 for g77 |
| xlf90 compiler | NAG f95 compiler |
| xlf compiler | g77 compiler |
| Abaqus | MPIch |
| Patran |
| Matlab |
| Maple |
| Auto |
| BLAS/BLACS |
| LINPACK/EISPACK |
| FFTPACK |
| Harwell Library |
| LAPACK |
| Lindo |
| MINPACK |
| NAG C Library |
| NPSOL |
| ODEPACK |
Please send comments and suggestions about this web page to
nic-support-l@lists.rpi.edu.
Last updated: 29 June 04