If you have requested and enabled your COE HPC account and would now like to get the most out of your use of the cluster, here are some helpful hints to optimize your experience:

Some software is already available to you through your default executable paths.  Other software is managed and made available using Lmod (lua-based environment modules).  Examples of software available via the modules system include Cmake, Cuda, compilers (e.g. GCC, LLVM, Intel, Nvidia), MPI, Python, and R.  For instance, suppose you have already reserved a GPU in Step 5 but now you need access to Cuda for your GPU programming.  To see available Cuda versions, do:

module avail cuda

To load the default version of Cuda, do:

module load cuda

You can confirm the version of Cuda you are using by:

nvcc --version

To view modules loaded in your environment, do:

module list

Check out this link for more examples and information on using Lmod to access and manage available software on the COE HPC cluster.

1. Home directory.

Every user with an active ENGR account has a fixed home directory quota of 25 GB. For those who need more space, please see additional storage options below.

2. HPC global scratch share directory (medium-term storage).

Those with a permanent HPC account (i.e. not a class account or other temporary HPC account but activated via the TEACH website) are given access to a global HPC scratch directory with a 1 TB quota that they can use to run their jobs from and to store temporary data to. The global scratch directory is located in /nfs/hpc/share/username, where your userame is your ONID, or OSU Network ID. This directory can be accessed through a shortcut called "hpc-share" located in your home directory, e.g.:

cd hpc-share

At present the HPC share directory is only visible on the HPC cluster (submit nodes and compute nodes). Be advised the HPC share is NOT backed up and should not be considered a place to store data long term, but a temporary place to store data generated by the HPC cluster. Also, there are no tools available to recover lost or deleted files from this filesystem. This share is subject to a purge policy of 90 days, where files older than 90 days may be purged depending on overall storage usage. Users will receive advanced notification before any files are purged. Important data should be backed up or copied to a longer term storage. 

3. Local scratch directories (short-term storage).

Every HPC compute node has a local scratch directory called /scratch. Most local scratch directories are only visible to the host compute node, but some GPU-heavy nodes with a large scratch space (> 1 TB) may be presented over the network and visible to the rest of the cluster in a directory called /nfs/hpc/{hostname} for ease of access. These local scratch directories can be useful for improving performance of workloads with heavy input/output requirements, or for just storing very short term data for a particular job. Files stored on a local scratch directory are not backed up, and may be purged anytime without warning, so be sure to back up any important data stored on a local scratch directory.

4. Other data storage options.

Project storage (Depot/Guille) and longer term storage (Attic) may be available to you through the College of Engineering. Also, there are multiple cloud storage options available through OSU. For more information on these storage options, click here.

Here are a few methods to transfer data to and from the COE cluster:

1) You can use the HPC portal for small data transfers to and from the COE HPC cluster (up to 25 GB max) - go to Files menu and select the "upload" or "download" buttons to transfer files.

2) Use a more robust and secure file transfer capable application like MobaXterm, WinSCP, Filezilla or Cyberduck. All of these are available for Windows, and Cyberduck and Fillezilla are available for MacOS, and Filezilla is available for Linux.

3) SFTP/SCP. If you are using Windows and MobaXterm for your ssh sessions, then you can open an sftp session to one of the submit nodes. 

If you are using a Mac or Linux, an alternative command line option is to open a terminal and use the sftp command, or scp to one of the submit nodes, e.g.:

sftp username@submit.hpc.engr.oregonstate.edu
-or-
scp myLocalFile username@submit.hpc.engr.oregonstate.edu:

File transfers are most easily accomplished when on campus or connected to the OSU VPN.

If you need to run a GUI-based application on the cluster, then you need an X11 server application installed on your computer.  If you are running Windows then this is already provided by MobaXterm, but if you are using Putty as your ssh client, then you will need to install Xming on your computer, and configure Putty to enable X11 forwarding.  If you are running macOS, then you need to install Xquartz, then log in using ssh with the "-Y" option to enable X11 forwarding, e.g.:

ssh -Y username@submit.hpc.engr.oregonstate.edu

then reserve resources using srun with X11 forwarding enabled:

srun {resource options} --x11 --pty bash

 

An alternative to using SSH or using X11 for GUI applications is to try out the OpenOnDemand HPC portal. You must be on campus or connected to OSU VPN (Step 3a) to access the portal. Currently available interactive sessions include "Basic Desktop", "Advanced Desktop", "Jupyter Notebook/Lab", "Matlab", "Mathematica", "RStudio", "Ansys", and "StarCCM+"!