Sockeye Introduction

Land acknowledgement

I would like to acknowledge that I work on the traditional, ancestral, and unceded territory of the Coast Salish Peoples, including the territories of the xwməθkwəy̓əm (Musqueam), Skwxwú7mesh (Squamish), Stó:lō and Səl̓ílwətaʔ/Selilwitulh (Tsleil- Waututh) Nations.

Traditional: Traditionally used and/or occupied by Musqueam people

Ancestral: Recognizes land that is handed down from generation to generation

Unceded: Refers to land that was not turned over to the Crown (government) by a treaty or other agreement

Setup VSCode

Checkout the instruction of your right platform below:

Connecting to Sockeye

ssh cwl@sockeye.arc.ubc.ca

Alternative way connecting to Sockeye

ssh-keygen -t rsa

Alternative way connecting to Sockeye

# Note these indent matters
Host  arc # or another alias you like
  HostName sockeye.arc.ubc.ca
  User cwl # replace your cwl

# arc is from our config
ssh-copy-id -i ~/.ssh/id_rsa.pub arc # or cwl@sockeye.arc.ubc.ca

# Need to replace cwl!
type $env:USERPROFILE\.ssh\id_rsa.pub | ssh cwl@sockeye.arc.ubc.ca "cat >> .ssh/authorized_keys"

Sockeye

# If you followed earlier configuration correctly
ssh arc # otherwise ssh cwl@sockeye.arc.ubc.ca

• Upon login, you land on the login node and spend most time here
• Use it to run large computational works
• Whenever you’re ready, you submit batch jobs to server
• You can log out and leave a job running
• Runs with PBS, changing to SLURM¹ soon

Directories on Sockeye

VScode in Sockeye

Using direct text editors like nano or vi/vim could be hard (steep learning curve)

So, use VSCode instead!

Accessing softwares

• Once you connect with ssh:
• There are no applications loaded
• You must tell system what you need
• Limited options, if need more try container or conda

module load bin1/x.y.z bin2/x.y.z # load two modules and their versions
module purge # unload all modules
module list # what you have loaded
module save name_col # save loaded modules to "<name_col>"
module savelist # list saved collections (<name col> this case)
module restore name_col # load saved collection
module avail # show all available modules
module spider mod_name # check available `<mod_name>` in sockeye

Example Job Script

#!/bin/bash
#PBS -l walltime=00:05:00,select=1:ncpus=2:mem=2gb
#PBS -N sample_job
#PBS -A st-singha53-1
#PBS -m abe
#PBS -M tliang19@student.ubc.ca
#PBS -o my_output.txt
#PBS -e my_error.txt

cd ${PBS_O_WORKDIR}
python hello_world.py

#!/bin/bash
#PBS -l walltime=12:00:00,select=1:ncpus=4:ngpus=2:mem=64gb
#PBS -N sample_job_gpu
#PBS -A st-singha53-1-gpu
#PBS -M tliang19@student.ubc.ca
#PBS -o my_output.txt
#PBS -e my_error.txt

# load modules
module load gcc apptainer cuda

cd ${PBS_O_WORKDIR}
python hello_world.py

#!/bin/bash
#PBS -l walltime=3:00:00,select=1:ncpus=4:mem=16gb
#PBS -I
#PBS -q interactive_cpu
#PBS -N sample_job
#PBS -A st-singha53-1

# Note here should be empty, since you will get back a bash shell 
# with the requested resources in the script.
# And this ones are recommended to use, 
# when you have simple experiments to conduct without GPU

#!/bin/bash
#PBS -l walltime=3:00:00,select=1:ncpus=4:ngpus=1:mem=16gb
#PBS -I
#PBS -q interactive_gpu
#PBS -N sample_job
#PBS -A st-singha53-1

# Note here should be empty, since you will get back a bash shell 
# with the requested resources in the script.
# And this ones are recommended to use, when you have simple experiments
# to conduct that uses GPU.

Submitting and other useful commands

• Suppose our pbs script is saved as job_script.sh

qsub job_script.sh # to submit job

qstat -u $USER # $USER is a pre-defined ENV variable

qdel some_job_id # some_job_id is from after submmision

qstat -x some_job_id # see resource comsumption of job

Summary workflow

1. Login via ssh
2. Modify your job script
   • required modules
   • required environment
   • required resources
3. Submit job
4. Inspect results
5. Pray and repeat until success!

Thanks!