Welcome to the WATGPU research cluster

Introduction

Welcome to WATGPU, a SCS owned GPU cluster aiming to facilitate access to computing resources for research purpose. This documentation serves as a comprehensive guide to understanding and utilizing WATGPU, a cluster managed through the Slurm workload manager.

Download pdf presentation here: 2024/07/25 version.

View the recording of the seminar from the 2024/07/25.

Getting access

Faculty (CS or cross appointed): Please contact watgpu-admin@lists.uwaterloo.ca
Student¹: With the agreement of the faculty you are working with, send an email to watgpu-admin@lists.uwaterloo.ca with them in CC.

Before making an account request, please load an ssh key at https://authman.uwaterloo.ca

Contact

If you require assistance while using WATGPU, you can contact the following:

Admin support: watgpu-admin@lists.uwaterloo.ca

Shared GPU Resources

WATGPU offers shared access to GPUs owned by the university and generous researchers. When these GPUs are idle, they contribute to a shared pool, providing users with enhanced computational capabilities.

Slurm: How it works

Slurm simplifies the user experience by allowing you to submit, monitor, and manage your computational jobs seamlessly. Through straightforward command-line interfaces, you can submit batch jobs, specify resource requirements, and monitor job progress. Slurm ensures fair resource allocation, allowing you to focus on your research without the complexity of manual resource management.

If you're new to shared GPU computing and Slurm, getting started with WATGPU is a breeze! Think of WATGPU as your personal computing powerhouse, shared with other users when their GPUs are idle. To begin, follow these simple steps:

Login: Access watgpu.cs using your credentials.
Submit a Job: Utilize the sbatch command to submit your script that you wish to run. Think of it like asking the server to perform specific computations for you using specific resources (like how many GPUs, how much memory ...).
Monitor Progress: Check the status of your job with the squeue command and see how it's progressing. You can use squeue to view the job queue and monitor job details.
Enjoy: Your job will be run by the server as soon as the requested resources are available.

For more in-depth information, visit this page.

Thank you for choosing WATGPU. We're here to simplify your computational tasks and enhance your research.

Happy Computing!

_{WatGPU cluster is a research resource. Access will only be granted to students actively involved in a SCS or cross appointed research group.}

Current cluster loads

The cluster consists of four servers. GPU utilization of the servers are shown below, and the links will take you to other dashboards detailing other performance metrics of the servers. Current loads are:

watgpu-100.cs

watgpu-200.cs

watgpu-300.cs

watgpu-400.cs

Managing virtual environments on the cluster

The default environment manager for all users on the WATGPU cluster is conda, though pip virtual environments are also supported.

You can create pip virtual environments while the base conda environment is activated, and after creating the pip virtual environment, you can deactivate the conda environment and activate the pip virtual environment normally.

(base) $ python -m venv <venv-name>
(base) $ conda deactivate
$ source <venv-name>/bin/activate

Installing useful tools through conda

Tools like nvcc and nvtop can be installed using conda since conda is also a package manager.

To perform a basic install of all CUDA Toolkit components using conda, run:

conda install cuda -c nvidia

You can install previous CUDA releases by following the instructions detailed in the Conda Installation section of NVIDIA's online documentation.

To install nvtop, which can be used to better monitor GPU utilization and GPU memory usage, run:

conda install --channel conda-forge nvtop

Further information about nvtop can be found here.

If a tool suggests you to use apt or apt-get to install it, please check if you can install the same tool through conda. If this isn't possible, you can send an email to watgpu-admin@lists.uwaterloo.ca and we can look into installing the tool for you.

SLURM

SLURM (formerly, Simple Linux Utility for Resource Management) is an application for managing tasks on computer systems.

The official SLURM cheatsheet can be found HERE

SLURM has two main operating modes: batch and interactive. Batch is the preferred mode for watgpu.cs: you can start your task and get an email when it is complete. If you require an interactive session (useful for debugging) with watgpu.cs hardware, see the salloc section.

Batch mode usage

You can submit jobs using an SLURM job script. Below is an example of a simple script: :warning: #SBATCH is the trigger word for slurm to take into account your arguments. If you wish to disable a line consider using ##SBATCH, do not remove # at the beginning of the lines.

#!/bin/bash
    
# To be submitted to the SLURM queue with the command:
# sbatch batch-submit.sh

# Set resource requirements: Queues are limited to seven day allocations
# Time format: HH:MM:SS
#SBATCH --time=00:15:00
#SBATCH --mem=10GB
#SBATCH --cpus-per-task=2
#SBATCH --gres=gpu:1

# Set output file destinations (optional)
# By default, output will appear in a file in the submission directory:
# slurm-$job_number.out
# This can be changed:
#SBATCH -o JOB%j.out # File to which STDOUT will be written
#SBATCH -e JOB%j-err.out # File to which STDERR will be written

# email notifications: Get email when your job starts, stops, fails, completes...
# Set email address
#SBATCH --mail-user=(email address where notifications are delivered to)
# Set types of notifications (from the options: BEGIN, END, FAIL, REQUEUE, ALL):
#SBATCH --mail-type=ALL
 
# Load up your conda environment
# Set up environment on watgpu.cs or in interactive session (use `source` keyword instead of `conda`)
source activate <env>
 
# Task to run
 
~/cuda-samples/Samples/5_Domain_Specific/nbody/nbody -benchmark -device=0 -numbodies=16777216

You can use SBATCH variables like --mem, for example the one above will assign 10GB of RAM to the job.

For CPU cores allocation, you can use --cpus-per-task , for example the one above will assign 4 cores to the job. The --gres=gpu:1 will assign 1x GPU to your job.

Running the script

To run the script, simply run sbatch your_script.sh on watgpu.cs

Interactive mode usage

You can book/reserve resources on the cluster using the salloc command. Below is an example:

salloc --gres=gpu:2 --cpus-per-task=4 --mem=16G --time=2:00:00

The example above will reserve 2 GPUs, 4 CPU cores, and 16GB of RAM for 2 hours. Once you run the command, it will output the name of the host like so:

salloc: Nodes watgpu308 are ready for job

here watgpu308 is the assigned host that the user can SSH to.

Ideally you want to run this command in either a screen or tmux session on watgpu.cs

Queues

To look at the queue of jobs currently, you can use squeue to display it.

The command scurrent will also give all the current job running on watgpu. It is useful to see which resources are in use at the moment.

Monitoring jobs

Current jobs

By default squeue will show all the jobs the scheduler is managing at the moment. It will run much faster if you ask only about your own jobs with

$ squeue -u $USER

You can show only running jobs, or only pending jobs:

$ squeue -u <username> -t RUNNING  
$ squeue -u <username> -t PENDING

You can show detailed information for a specific job with scontrol:

$ scontrol show job -dd <jobid>

Do not run squeue from a script or program at high frequency, e.g., every few seconds. Responding to squeue adds load to Slurm, and may interfere with its performance or correct operation.

Cancelling jobs

Use scancel with the job ID to cancel a job:

$ scancel <jobid>

You can also use it to cancel all your jobs, or all your pending jobs:

$ scancel -u $USER

$ scancel -t PENDING -u $USER

Queue and Priority

Jobs in WATGPU are ran depending on their priority. Higher priority jobs go faster in the queue and can sometimes requeue running jobs to free resources for themselves.

To work with this priority, WATGPU resources have been separated into several partitions. The partition on which a job is running defines its possible allocated resources and its priority. Selecting a partition to run a job is easy:

For interactive sessions: simply add --partition=<PARTITION> argument in the salloc command.
For batch jobs: add #SBATCH --partition=<PARTITION> in your .sh script

Following is a presentation of the different partitions available in WATGPU:

ALL

This is the default partition. A job submitted to ALL will be able to run on any available resources on WATGPU. Jobs will have standard priority for the queue and might be preempted and requeued if a higher priority job is in need of resources.

All users have access to this partition:

#SBATCH --partition=ALL

SCHOOL

A job submitted to SCHOOL will be allocated only to GPUs owned by the School. While fewer resources are available in SCHOOL than in ALL, the chances of getting preempted by higher priority jobs are reduced. Jobs will have standard priority for the queue and might be preempted and requeued if a higher priority job is in need of resources.

All users within the School have access to this partition:

#SBATCH --partition=SCHOOL

<GROUP>

When a group <GROUP> has entrusted their GPU(s) to WATGPU, they have access to the partition named <GROUP> linked to their contributed GPU(s). Jobs will have high priority for the queue and will preempt and requeue lower priority jobs (from ALL and SCHOOL) if there are not enough resources available with a limit of GPU(s) up to their contribution.

All users from a group have access to their group partition:

#SBATCH --partition=<GROUP>

Users using their group's partition to request GPUs must now specify the group name for the GPU request in their jobs:

#SBATCH --gres=gpu:<GROUP_name>gpu:1

Accessing Jupyter notebooks in the cluster

One can access jupyter notebooks on compute systems using SOCKS proxy and ssh through watgpu.cs. From a (Linux/OSX) terminal ssh to watgpu.cs:

ssh -D 7070 user@watgpu.cs.uwaterloo.ca

On a browser on your local (client) system, configure traffic to use a SOCKS proxy at localhost port 7070. FoxyProxy for Firefox can make this configuration easy to set up and modify.

Preliminaries: Create a conda environment that includes the jupyter server

Also add required conda packages for your working environment e.g. pytorch:

conda create -y -n jupyter-server
conda activate jupyter-server
conda install -c conda-forge pytorch-gpu
pip install jupyter
conda deactivate

Using jupyter notebooks with your environment

Make an interactive reservation with the SLURM scheduler:

salloc --gres=gpu:1 --cpus-per-task=4 --mem=32G --time=1:00:00

Once the reservation starts, ssh to the allocated compute system e.g. watgpu208:

ssh watgpu208

Activate your jupyter-server environment and start a jupyter notebook:

conda activate jupyter-server

jupyter notebook --ip $(hostname -I | awk '{print $1}') --no-browser
...
[I 2023-10-13 02:18:02.902 ServerApp] notebook | extension was successfully loaded.
[I 2023-10-13 02:18:02.902 ServerApp] Serving notebooks from local directory: /u3/ldpaniak
[I 2023-10-13 02:18:02.903 ServerApp] Jupyter Server 2.7.3 is running at:
[I 2023-10-13 02:18:02.903 ServerApp] http://192.168.152.121:8888/tree?token=141a606c1ec9d9f76f65395bd1a4042fb3a5e04307283592
[I 2023-10-13 02:18:02.903 ServerApp]     http://127.0.0.1:8888/tree?token=141a606c1ec9d9f76f65395bd1a4042fb3a5e04307283592
[I 2023-10-13 02:18:02.903 ServerApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[C 2023-10-13 02:18:02.915 ServerApp] 

To access the server, open this file in a browser:
    file:///u3/ldpaniak/.local/share/jupyter/runtime/jpserver-232011-open.html
Or copy and paste one of these URLs:
    http://192.168.152.121:8888/tree?token=141a606c1ec9d9f76f65395bd1a4042fb3a5e04307283592
    http://127.0.0.1:8888/tree?token=141a606c1ec9d9f76f65395bd1a4042fb3a5e04307283592

Copy the link with the 8888 port to your broswer which has been configured to use SOCKSv4 proxy on localhost port 7070. Your juptyer notebook will be available.

Be sure to shut down the server when done with Control-c.

Custom kernels

There is a single default kernel at the moment: "Python 3". You can also create your own kernels by opening a Terminal inside the notebook:

Once you've opened the terminal you can create your own kernel. Below is an example:

conda create --name myenv # create a custom conda environment which will install packages to, and add to the notebook as a kernel

conda install --yes numpy # install a package you want

conda install -c anaconda ipykernel #install ipykernel which we will use to add kernel to notebook

python -m ipykernel install --user --name=myenv # add the conda environment as a kernel

VSCode Tutorial: Working with WATGPU

In this tutorial, we'll walk you through the process of setting up and using Visual Studio Code (VSCode) to work with WATGPU. This guide will cover connecting to the login gateway and accessing specific clusters (e.g., watgpu108, watgpu208, ...) after allocating resources.

Note that this method of connecting to WATGPU should be used for debugging and understanding your code via testing and notebooks. If you wish to run long experiments, please use the SBATCH command.

Prerequisites

Before you begin, make sure you have the following prerequisites installed:

Quick steps to install the Remote - SSH extension:

Open Visual Studio Code
Click on the Extensions icon in the Activity Bar on the side of the window.
Search for "Remote - SSH" in the Extensions view search box.
Install the "Remote - SSH" extension.

Step 1: Configuring VSCode

Once the extension is installed, follow these steps to connect to the login gateway (replace your_username with your actual username):

Press Ctrl + Shift + P (Windows/Linux) or Cmd + Shift + P (Mac) to open the command palette.
Type "Remote-SSH: Open SSH Configuration File ...", select it and select the desired file (recommended file: ~/.ssh/config).

Insert the following part and replace <username> and <path/to/private/key> with your username and the path to your private key registered in WatGPU:

#WATGPU
Host watGPU
    User <username>
    IdentityFile <path/to/private/key>
    HostName watgpu.cs.uwaterloo.ca

Host watGPU108
    User <username>
    IdentityFile <path/to/private/key>
    HostName watgpu108
    ProxyJump watGPU

Host watGPU208
    User <username>
    IdentityFile <path/to/private/key>
    HostName watgpu208
    ProxyJump watGPU

Host watGPU308
    User <username>
    IdentityFile <path/to/private/key>
    HostName watgpu308
    ProxyJump watGPU

Save and refresh the Remote - SSH extension menu.

Step 2: Allocate Resources and Connect to Cluster

After successfully connecting via ssh to watgpu.cs.uwaterloo.ca, you'll need to allocate resources using salloc and then connect to a specific cluster:

Follow the Interactive mode part of the documentation to obtain resource allocation.
Once resources are allocated, connect to the appropriate cluster in VSCode:

That's it! You are now set up to work with WatGPU using Visual Studio Code.