SSH Access

I've created some aliases for the Otter machines that you're welcome to use.

ssh_surrey() {
  [[ -z "$1" ]] && echo "usage: ssh_surrey <server>" && return 1
  local host="$1"

  case "$host" in
    aisurrey)   host="aisurrey-submit01" ;;
    otter[0-9][0-9][0-9]) host="$host.eps" ;;
  esac

  sshpass -p "$(op read op://Private/MySurrey/password)" ssh "tl01159@$host.surrey.ac.uk"
}
alias ai_surrey='ssh_surrey aisurrey'
alias eureka2='ssh_surrey eureka2'
alias otter111='ssh_surrey otter111'
alias otter112='ssh_surrey otter112'
alias otter113='ssh_surrey otter113'
alias otter114='ssh_surrey otter114'
alias otter115='ssh_surrey otter115'

Install Miniconda

wget <https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh>
bash ~/Miniconda3-latest-Linux-x86_64.sh

There should be an environment.yaml file provided by @xiaoguang Liang.

You should be able to create a conda environment by doing

conda create -f environment.yaml
conda activate faice

tmux

<aside> ❗

You can run multiple SSH and tmux sessions simultaneously. However, be mindful of your storage usage and remember to regularly clear your W&B local logs. As a best practice, clear your wandb/ local logs before each run.

Alternatively, you can point your wandb artifacts and generated files to a volume with bigger storage like scartch. For a list of env vars, you can consult here.

export WANDB_DIR="<YOUR_PATH_HERE>"
export WANDB_ARTIFACT_DIR="<YOUR_PATH_HERE>"

If you're doing research with Simon, you can request a 200GB storage space in vol/research/<your_student_id>.

You can check your disk usage by:

df -h /vol/research/<your_student_id>
Filesystem                                                                    Size  Used Avail Use% Mounted on
isilon01-az3.surrey.ac.uk:/ifs/isilon01/az3/Research/RS501/<your_student_id>  200G   30G  171G  15% /vol/research/<your_stdudent_id>

</aside>

Make sure you use tmux, otherwise your SSH session will be terminated if you disconnect or the connection drops. tmux allows you to keep your sessions running in the background and reattach to them later. This is especially important for long-running experiments and processes.

tmux new -t <experiment_name>
conda activate faice
# run your experiment command after this.
# press ctrl + B and then D to detach a session

You can then check the status of your runs on W&B. To re-attach a session

tmux attach -t <experiment_name>