Insomnia - Software

When you login to Insomnia, you land on one of the 2 login nodes. As a general rule, you should do your actual work on a compute node. Meaning when you login, you should first execute a line like this:

srun --pty -t 0-02:00 -A <ACCOUNT> /bin/bash.              (CPU compute node)
srun --pty -t 0-02:00 --gres=gpu:1 -A <ACCOUNT> /bin/bash. (GPU compute node)

(The above lines will open a session on a random compute node for a maximum of 2 hours. You can lengthen or shorten the time as you please. The absolute upper limit is 5 days on a node your group owns. 12 hours on another group's node.)

This is important because all of the software noted in the below table that is loaded automatically is only seen on a compute node. It will not be seen on a login node.

Use the following command to verify whether software/packages can be found on Insomnia:

$ module avail

NOTE: For a good guide on how to use environment modules to easily load your software environment, please see:

https://lmod.readthedocs.io/en/latest/010_user.html

Table of installed Software

The table below shows software already installed  on the cluster as a module. NOTE: On the Insomnia cluster, some software is automatically available and need not be manually loaded. Such software is noted.

The list may be partial and not totally up-to-date at any given time. Requests for software can be submitted for consideration through an emailed request to hpc-support@columbia.edu

Mathematica

The first time you launch Mathematica you will need to provide the host details of the MathLM (license) server. Using the Activating Mathematica guide, click 'Other ways to activate' then choose 'Connect to a Network License Server' and enter the IP address, 128.59.30.140

OpenMPI Settings  *

The default OpenMPI on Insomnia is  openmpi-5.0.2, which is provided by Nvidia Mellanox and optimized for the MOFED stack. You will receive the following warnings when using mpirun/mpiexec:

WARNING: There was an error initializing an OpenFabrics device.

  Local host:   g###
  Local device: mlx5_0

Default device parameters will be used, which may result in lower performance. You can edit any of the files specified by the
btl_openib_device_param_files MCA parameter to set values for your device.

NOTE: You can turn off this warning by setting the MCA parameter btl_open ib_warn_no_device_params_found to 0.

You can pass the following option, which will use ucx which is default as of version 3.x:

--mca pml ucx --mca btl '^openib' 

To help with the following warning:

Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages

You can also add:

--mca orte_base_help_aggregate 0

If you choose to use the openmpi/gcc/64/4.1.1_cuda_11.0.3_aware module, this version expects a GPU and will throw the following warning on non-GPU nodes:

The library attempted to open the following supporting CUDA libraries, but each of them failed. CUDA-aware support is disabled.
libcuda.so.1: cannot open shared object file: No such file or directory
libcuda.dylib: cannot open shared object file: No such file or directory
/usr/lib64/libcuda.so.1: cannot open shared object file: No such file or directory
/usr/lib64/libcuda.dylib: cannot open shared object file: No such file or directory

If you are not interested in CUDA-aware support, then run with --mca opal_warn_on_missing_libcuda 0 to suppress this message. If you are interested in CUDA-aware support, then try setting LD_LIBRARY_PATH to the location of libcuda.so.1 to get passed this issue.

You can pass this option:

--mca opal_warn_on_missing_libcuda 0

RStudio in an Apptainer container

On Insomnia, rstudio can be loaded by leveraging an interactive session, i.e., srun, and using a Apptainer container via the Rocker Project which provides containers from where this tutorial comes . First you will need to download the Apptainer image. This will require two terminal sessions. One session will be used to connect to a compute host and run rstudio. The other terminal session will be used to initiate SSH Port Forwarding/Tunneling to access the resource. The first step is to request an interactive session so you can run rstudio server from a compute node as follows. Please be sure to change your group as noted by "-A".

srun -X -A <GROUP> --pty -t 0-01:00 -X /bin/bash

Apptainer is loaded automatically and no longer a separate module the way Singularity was on previous clusters. We can go ahead and use it to pull the container. This will take a few minutes.

On subsequent sessions you skip this step as the container will remina.

apptainer pull --name rstudio.simg docker://rocker/rstudio:4.3.1

In order for RStudio to start in a browser via an interactive session you will need the IP address of the compute node. Note that the IP below will likely be different for you:

$ hostname -i 
10.197.16.39           (REMEMBER, this is only an example IP. Yours will likely be different)

From RStudio 4.2 and later, some added security features require binding of a locally created database file to the database in the container. Don't forget to change the password.

mkdir -p run var-lib-rstudio-server

printf 'provider=sqlite\ndirectory=/var/lib/rstudio-server\n' > database.conf

PASSWORD='CHANGEME' singularity exec \
   --bind run:/run,var-lib-rstudio-server:/var/lib/rstudio-server,database.conf:/etc/rstudio/database.conf \
   rstudio.simg \
   /usr/lib/rstudio-server/bin/rserver --auth-none=0 --auth-pam-helper-path=pam-helper --server-user=$USER

This will run rserver in a Singularity container.

Now open another Terminal and start the RStudio rserver session using Port Forwarding to connect a local port on your computer to a remote one on Insomnia.

ssh -Nf -L 8787:10.197.16.39:8787 myUNI@som.rcs.columbia.edu           (NOTE: THIS IS NOT FOR WINDOWS. On a PC, see below)

The above Port Forwarding line works on Linux/Unix/MacOS, or any virtual machine running Linux. For Windows, we are assuming you are using PuTTY and the above line would be accomplished using the alternative steps below:

I.    Launch PuTTY

II.   In the "Session" category (the default screen when you open PuTTY), enter som.rcs.columbia.edu in the "Host Name (or IP address)" field. 

III.   Ensure the "Port" field is set to 22, which is the default SSH port.       Optionally, you can enter a name in the "Saved Sessions" field and click "Save" to save these settings for future re-use

IV.   In the "Category" pane on the left, expand the "SSH" category by clicking the + sign next to it.

V.    Click on "Tunnels" under the "SSH" category.

VI.   In the "Source port" field, enter 8787.

VII.   In the "Destination" field, enter 10.197.16.39:8787.    (Remember,  10.197.16.39 is only an example IP, your actual one may be different)

VIII.   Click the "Add" button. You should see the forwarded port appear in the list box as L8787 10.197.16.39:8787.

IX.     Optional but recommended,   To save this configuration for future use, go back to the "Session" category, and enter a name for the session in the "Saved Sessions" field (e.g., <myuni>@som.rcs.columbia.edu) and click the "Save" button.

X.      Click the "Open" button at the bottom of the PuTTY window and connect to som.rcs.columbia.edu

Next, use a web browser to connect to to your locally forwarded port.

In any browser, e.g., Google Chrome, Firefox, Edge, etc. etc., enter localhost:8787 in the web address field. The login screen for RStudio will appear and you enter your UNI (without the @columbia.edu) and whatever password you put from the above example (where it says CHANGME). If you need a session that last longer than an hour you can add --auth-timeout-minutes=0 --auth-stay-signed-in-days=30 to the end of the above singularity. command. A sample Slurm bash script is available at Rocker Project's Singularity page. If you run into an error such as [rserver] ERROR system error 98 (Address already in use), you can specify the port number as an additional option, e.g.,  --www-port 8788. The new command could look like this:

mkdir -p run var-lib-rstudio-server

printf 'provider=sqlite\ndirectory=/var/lib/rstudio-server\n' > database.conf

PASSWORD='CHANGEME' singularity exec \
   --bind run:/run,var-lib-rstudio-server:/var/lib/rstudio-server,database.conf:/etc/rstudio/database.conf \
   rstudio.simg \
   /usr/lib/rstudio-server/bin/rserver --www-port 8788 --auth-none=0 --auth-pam-helper-path=pam-helper --server-user=$USER

***IMPORTANT NOTE***

When you are finished with RStudio remember to stop the Singularity session via ctl-c, and you can kill the SSH forwarding session by finding the process ID (PID) with the following command, noting that the PID is the second column:

ps -ef | grep 8787
  503 82438     1   0 10:56AM ??         0:00.31 ssh -Nf -L 8787:10.197.16.39:8787 myUNI@som.rcs.columbia.edu
kill -9 82438

Visual Studio Code Server

Visual Studio Code is an Integrated Development Environment that some like to use on their laptops. If you are familiar with that, the HPC has a server-side version hosted on the compute nodes (NOT the login nodes) for users to create code directly on Insomnia with access to their files. To use it, do the following:

After logging into Insomnia, start an srun session to a CPU or GPU node of your choice:

Then type

$ code tunnel

You will get a message like the one below:

A Github account is not required. Simply open a browser window and point it to the device login line in the last sentence. The browser window will ask for the unique code the line provides.

After that, you will get another custom URL to load a VS Code editing window hosted from the server.