This page contains some core skills a user must develop after gaining access to our HPC system.

To connect to an HPC System

To access an HPC system via SSH-2 use:
PuTTY, OpenSSH, or SecureCRT; Example:

or

To transfer files to and from the gird using SFTP use:
WinSCP, OpenSSH, or SecureFX; Example:

Architecture

64bit code and 32bit code are both supported; what is optimal varies on a case by case basis. M2070 GPUs may be available. CPU architecture, CPU speed, memory, and networking may also be important factors. Use the list of queues (qstat -q) while making decisions about where to run jobs, especially if you want to make use of GPU resourses.

$HOME & $WORK

When a job script is submitted PBS creates a shell to run the job on the assigned nodes. User $HOME directories are located on a network fiel system and are accessible from all of the nodes. Queued job should run in a directory on the assigned node which is part of the parallel file system, identified by the environment variable $WORK.

To find cluster status and job summary

Modules commands examples

To start an interactive job

On Euclid

On Cy-Tera

-I starts an interactive job
-q specifies a queue (cpuq or gpuq for Euclid and cpui or gpui for Cy-Tera)
-l number of nodes and cores per node needed by a job

This will open a shell on the first node assigned. This is useful for jobs that have graphical interfaces or setting up job scripts.

Example MPI code: hello.c

To compile a parallel job

To create a job script

A job script is a list of instructions that tell PBS how to run your job. You can modify a job script in a text editor such as Vim, nano, joe, pico, emacs etc. You should do such work on the head node.

On Euclid

On Cy-Tera

To submit a job

On Euclid

On Cy-Tera

Makefile Example

The examples mentioned hereby can be found in euclid: /opt/examples/coreskills