UBC ARC Sockeye Documentation

Use vscode to login

• Open a vscode interface, press Ctrl+Shift+P (Command+Shift+P for Mac) to open the search bar.
• In the search bar, search remote ssh, then click on new SSH connection
• Type <cwl-account>@sockeye.ubc.ca and save the ssh config
• Connect to sockeye with your CWL password and DUO verification.

Install and load software

See the available software:

module avail

Load software:

module load gcc
module load cuda
module load miniconda3

Create your conda environment

In a terminal,

module load miniconda3
module load git
source ~/.bashrc

Note that some packages may require git, so also load git

Create a conda environment from yml file:

conda env create -f environment.yml

It is the first time to activate conda env under your account,

conda init bash

Then open a new terminal,

conda activate <your conda env>

Run a GPU job

Create a bash file called run_job.sh:

#!/bin/bash
 
#SBATCH --job-name=jobname            
#SBATCH --account=alloc-code-gpu    
#SBATCH --nodes=1                  
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=12                           
#SBATCH --mem=64G                  
#SBATCH --time=72:00:00             
#SBATCH --gpus-per-node=2
#SBATCH --output=output.txt         
#SBATCH --error=error.txt          
#SBATCH --mail-user=your.name@ubc.ca
#SBATCH --mail-type=ALL                               

module load gcc
module load cuda
module load <software_package_1>
module load <software_package_2>
 
cd $SLURM_SUBMIT_DIR

<gpu_executable>

Submit run_job.sh by:

sbatch run_job.sh

You can check the running or pending jobs by running this command:

squeue -u <your_CWL> -t RUNNING
squeue -u <your_CWL> -t PENDING

Data Transfer

To download the data from sockeye, use the following command:

scp -r cwl@sockeye.arc.ubc.ca:path local_pat

Reference

• Official Documentation https://confluence.it.ubc.ca/display/UARC/GPU+Jobs+in+Detail