Skip to main content Link Menu Expand (external link) Document Search Copy Copied

Digital Alliance of Canada

We make use of the Digital Alliance of Canada servers (specifically the Graham server) for nearly all of our computing and storage needs. This is a specialized system with mainly command-line access and submission of jobs for computing. Through it, we have access to hundreds of CPUs, many GPUs, and hundreds of TB of storage. These tools allow you to access servers supported by the Canadian government for research purposes, which are useful for jobs that are very computationally heavy.

You will need to sign up for an account to access this resource - see below for more information.

Table of Contents

Getting an account

Create an account at CCDB.

Make sure you include your Sponsor, Ali Khan (CCRI: xkc-513-04). Once the sponsorship request has been accepted, you will be granted privleges to actually use SHARCNET.

To log in to the system, use ssh <user>@graham.computecanada, replacing <user> with your SHARCNET user name in the terminal.

Passwordless SSH

If you are using sshfs, you should always have passwordless ssh set-up, otherwise you may encounter issues with your IP becoming blacklisted due to too many failed reconnection attempts.

If you are using the CBS server, there is an automated process to set this up by running setup_sshfs_alias. If not, instructions to set this up can be found below (commands should be entered into a terminal):

  1. Generate a ssh keyfile:
     ssh-keygen -t rsa
    
  2. Copy this to the remote server
     ssh-copy-id -i ~/.ssh/id_rsa.pub <user>@graham.computecanada.ca
    

Mounting Graham data locally with sshfs

Mac/Linux

After setting up the passwordless SSH, follow the next steps:

  1. Create a new folder in the location where you wish to mount your Graham home folder. For example: mkdir -p ~/graham
  2. Create a new folder in your local home folder: mkdir -p ~/bin.
  3. Create a new file in the previously created folder: code ~/bin/graham. When VScode opens, put the following content in the file and save:
     #!/bin/bash
     mount_dir="<MOUNT_DIR>"
     if (! mountpoint -q $mount_dir); then
         sshfs <user>@graham.computecanada.ca:/home/<user>/ $mount_dir -o ServerAliveInterval=15,ServerAliveCountMax=3,Compression=no,follow_symlinks
     else
         umount $mount_dir
     fi
    

    Replace <MOUNT_DIR> with the directory created in the first step. Also, substitute <user> with your corresponding user.

  4. Add execution permissions to it. You can use chmod +x ~/bin/graham. Here, the +x part of the command adds execution permission to the file.
  5. Export to path permanently. Open your bashrc file using code ~/.bashrc. Add the following line to the end of the file:
     export PATH="~/bin:$PATH"
    
  6. Reload your bashrc file: source ~/.bashrc.

Now, you should be able to mount or unmount your Graham home folder by running graham in the terminal.

If something like the following prints after typing code <somefile>:

command not found: code

You may need to install VScode to your system first. See here for installation instructions

Windows

For Windows, you can follow the instructions put together here

File transfer with Globus

The recommended way of transferring large amounts of data from your local filesystem and Graham is to use Globus. To do so, follow the instructions below:

  1. Log on to Globus
  2. Click “Endpoints” tab and subsequently subtab “administered by me”. If no personal endpoints, add “Globus Connect Personal endpoint”. Set a Display Name and download the appropriate application for your operating system (note: this is available through CBS).
  3. Once successful, go to the “Transfer Files” tab and set Endpoints depending on the direction of transfer.

For transfer to/from Graham, use computecanada#graham-dtn

Setting group read/write permissions

Run these commands the first time you login to a new cluster system to grant group read/write to project folders (to allow easier debugging):

chgrp -R ctb-akhanf ~/projects/*/$USER;
chmod -R g+rwXs ~/projects/*/$USER;

To grant access to your home directory:

chgrp -R ctb-akhanf $HOME;
chmod -R g+rwXs $HOME;

Running jobs

Neuroglia-helpers (legacy)

The neuroglia-helpers scripts provide an easier interface to submit jobs, including a BIDS-app job submission script (bidsBatch).

To install the neuroglia-helpers scripts, see the README here.

kslurm

Newer helper and wrappers for running commands on Graham have been developed in the utility package kslurm.

To install kslurm, see the README here.

Job-templates

It is importantto understand compute node specs and appropriately allocate resources using some combination of parameters like --mem, --cpus-per-task, --ntasks, etc. Node types and characteristics can be found here.

The neuroglia-helpers and kslurm job templates are optimized for the size of nodes and our allocation on Graham. Some examples are found below:

  • Regular: 8cores/32GB/24hr (1/4 of a node)
  • Short: 8cores/32GB/3hr (1/32 of a node)
  • Fat: 32cores/128GB/24hours (1 node)

Useful SLURM Commands

Here are some useful commands to help monitor usage on Graham.

  • sq - See what is in the queue for your user (running / waiting)
  • sshare -U $USER - See share usage
  • sshare | grep akhanf - See share usage for entire lab
  • scancel <jobid> - Remove job associated with <jobid>
  • scancel -u $USER - Remove all your submitted jobs (BE CAREFUL)
  • sacct -j <jobid> --format JobID,ReqMem,MaxRSS,Timelimit,Elapsed - Check completed job ussage

Run the following to set-up your .bashrc to enable a more readable sacct output:

echo “export SLURM_TIME_FORMAT=relative” >> ~/.bashrc
echo “export SACCT_FORMAT=JobID%-20,Start%-10,Elapsed%-10,State,AllocCPUS%8,MaxRSS,NodeList” >> ~/.bashrc

Visualization

Login node

You can also use X forwarding to visualize your data via SSH. To do so, run the following command:

ssh -Y <user>@graham.computecanada.ca

By default, this brings you to the login node which is not a compute node. This should only be used for viewing images.

Compute nodes

In Graham, it is possible to run a VNC server in a compute node, which allows the visualization of results, such as plots, or using GUI applications as MATLAB. Follow these instructions to get a VNC connection to a compute node in Graham (based on the information found here):

  1. Download TigerVNC from the official download page and install it. Refer to the instructions here for more details.
  2. Use ssh to connect to Graham login node.

    It is recommended to install neuroglia-helpers or kslurm if not already done. This will facilitate the request of resources in Graham.

    1. Request an interactive session with the needed resources. This can be done using neuroglia-helpers or kslurm.
    2. Once the interactive session is opened, take note of the node to which you are connected and then run:
       export XDG_RUNTIME_DIR=${SLURM_TMPDIR}
      
    3. Run vncserver.
    4. Finally, run the following command to know the connection port of the server:
       grep /home/<username>/.vnc/<log_file>.log
      

      This path will be shown in the terminal after executing vncserver in previous step.

    If you are configuring the VNC connection for the first time, make sure that you DO NOT leave the password empty. Save this password somewhere safe.

  3. In a new terminal, run the following command to set the ssh tunnel between your computer and the server configured in the previous step.
     ssh <username>@graham.computecanada.ca -L <port on your computer>:<allocated node>:<port from the node> 
    

    Fill in the corresponding information. For the port on your computer, you can use the 5902. In <allocated node>, put the node that you allocated after step 3, starting with gra followed by a few numbers. Fill <port from the node> input the port from step 2.e.

  4. Finally, open your VNC viewer application and connect to the local port that you decided in the previous step (for example: localhost:5902) and click on ‘Connect’. Use the password that you configured in the step 2.d.