I renewed my account with ComputeCanada, which is available for all Canadian University researchers (professors and adjuncts), and students and postdocs they work with. Here are a few notes that might be handy later.

Accounts

Sign up for an account.

This will get you access to the clusters

Login

Login with ssh:

ssh username@beluga.computecanada.ca

Use the same password as the one for your Compute Canada account here.

You can use SSH keys to login without a password. After you’ve followed the instructions at that link, you won’t have to use your password anymore.

For a little extra convenience, especially if you use multiple servers with different keys and usernames, add your details to .ssh/config:

Host beluga
     HostName beluga.computecanada.ca
     User username
     IdentityFile ~/.ssh/my_rsa_key

This will allow me to login to the beluga.computecanada.ca cluster as user username, using the private key ~/.ssh/my_rsa_key, with just the following command (and no password):

ssh beluga

This also allows me to open files on the cluster directly from my local Emacs instance:

M-x find-file
/ssh:beluga:~/path/to/file/on/beluga

Working with R

Modules

After a long day spent diving through dependencies, I found the following minimal set of modules necessary to install the packages used our GBS analysis:

module load nixpkgs/16.09 gcc/7.3.0 gdal/3.0.1 udunits/2.2.26 proj4-fortran/1.0 proj/6.3.0 r/4.0.2

You run this command (on a single line) at the terminal after logging into the cluster. After you’ve loaded the modules, you can save the collection for future sessions with:

module save r_modules

On our next, and all subsequent sessions, we can load R into an interactive session via:

module restore r_modules

Note that you may need to add additional modules to support new packages you install. Sometimes install.packages will give you an informative error message. Sometimes you’re not so lucky. Check to see which actual package causes the error (i.e., the one you’re installing, or one of its dependencies), and check the source repository for the problematic package for additional clues.

Shared R library

I set up a shared drive for everyone in the lab to use. This will save a lot of time, as installing packages from source on Linux can take a while.

mkdir $HOME/projects/def-tsmith/lab
setfacl -d -m g::rwx $HOME/projects/def-tsmith/lab

def-tsmith is my group name, if you don’t work with me you’ll need to use your supervisor’s project name instead.

To access this library, after you have loaded the R modules, execute the following command:

export R_LIBS=~/projects/def-tsmith/lab/R/x86_64-pc-linux-gnu-library/${EBVERSIONR%.*}/

EBVERSIONR is the version of R installed, including the bugfix release number (e.g., 4.0.2). We only use the major.minor numbers for the library directories (4.0), so I’ve trimmed off the last number via parameter substitution.

After you’ve done this, you can start R in an interactive session via R. You’ll have access to all the packages I’ve already installed, and any new packages will be installed into the same shared library.

Jobs

To load the modules and access the shared library from your scripts, use the following:

#!/bin/bash
##
##SBATCH settings for the job
##...

module restore r_modules
export R_LIBS=~/projects/def-tsmith/lab/R/x86_64-pc-linux-gnu-library/${EBVERSIONR%.*}/

## 'computation.R' contains the code you want to run:
Rscript computation.R