Logging in¶
To remotely access Hyak:
ssh [UW Net ID]@klone.hyak.uw.edu
You’ll be prompted to authenticate.
Starting a session¶
In order to load miniconda and do other intensive tasks, you will need to start an interactive session. For example:
srun -p gpu-a40 -A escience --nodes=1 --ntasks-per-node=4 --time=2:00:00 --pty /bin/bash
More information on flags can be found here
Managing your disk space¶
The home directory you begin in after logging in has limited space. More (but not persistent) space can be accessed by moving large files, caches, and working directories to /gscratch. For example:
cd /gscratch/escience/
mkdir [UW Net ID]
Any command with a cache or working directory flag should probably be pointed to a location in /gscratch, or a memory error may result.
Setting up conda¶
For many of the tasks that we do, we will want to use conda for environment
management. Because our home folders on hyak are very limited in terms of
disk space, it is best to create the environments outside of our home folders.
This can be done using the -p flag:
conda create -p /gscratch/escience/<netID>/<env name>After creating it for the first time, you will want to activate the env every time you install packages and work with them.
Loading modules¶
Once resources have been allocated and the session begins, you can perform computational tasks. For example:
conda activate /gscratch/escience/<netID>/<env name>
python my_script.py
You may want to load modules that give you access to specific programs, such as
module load apptainer. Conda does not require a module and runs as soon as
you log in.
More information on modules can be found here
Submitting jobs to slurm¶
You can also submit a task to be queued and executed through slurm.
The status of these jobs, as well as interactive sessions, can be seen with:
squeue -u [UW Net ID].