Get started with TIR Parallel File System

Create a Parallel File System (PFS) and mount it on a TIR Instance at a path you choose. PFS is available in Delhi, so work in a TIR project there with an instance already running. To create one, see Get started with TIR Instances first.

Create the PFS

1. In TIR, open your Delhi project. In the side menu, under Storage, select Parallel File System.
2. Select Create Parallel File-System.
3. In Parallel File-System Name, type a name. It must start with a lowercase letter, end with a letter or digit, use only lowercase letters, digits, and hyphens, and be at most 30 characters.
4. Set PFS Size to the number of TB you need, from 1 TB up to your account’s limit, 250 TB by default. Use the slider or type the number.
5. Select Create.

The new PFS shows Pending while TIR sets it up, then initializing. It reaches OK and is ready to mount in about 80 seconds for 1 TB. TIR emails you when it reaches OK. Billing starts at that point and runs until you delete the PFS; see How is a PFS billed?.

Mount the PFS on an instance

Mount the PFS once it reads OK.

1. Open your instance’s page and select the Parallel File-System tab.
2. In the PFS’s row, type a mount path.
3. Select Mount. Mounting restarts the instance, so the Mounting PFS dialog asks you to confirm. Tick the box, then select Continue.

The instance restarts, and the PFS is then available at the mount path you typed.

Check the mount

Once the instance is Running again, open a terminal in JupyterLab or over SSH and list the mount path, replacing /mnt/pfs with the path you typed:

df -h /mnt/pfs

The output shows the PFS with its size. Anything you write there is now shared with every other instance or cluster that mounts the same PFS. For the other mount choices, including read-only, see Mount a PFS on an instance. For help with any mount message, see Troubleshooting TIR Parallel File System.

What to do next

• Create a PFS explains every choice on the create form, including asking for a higher size limit.
• Mount a PFS on a training cluster to use a PFS from a training cluster or a Slurm cluster instead of an instance.
• Increase the size of a PFS when you need more space.