Skip to main content

Before you start

You need four things before you create a cluster: a balance, a shared filesystem, a VPC and an SSH key for each person. The filesystem and the VPC must be in the India-02 region.

Balance​

Your balance must cover one hour of the whole cluster. If it does not, the create page shows this message and Create stays off:

Your balance doesn’t cover one hour of this cluster. Add balance and try again.

If your balance is zero, the button shows Add credits to continue instead.

The create page shows the price per hour before you create. Read it there. To add balance, go to Recharge. Refer to Billing for what Jarvislabs charges and when.

Shared filesystem​

A cluster needs one shared filesystem. It holds each person's home folder, and it stays when you delete the cluster.

  1. Go to the Shared Filesystems page.
  2. Create a filesystem in the India-02 region.

Refer to Shared Filesystems for the steps and sizes.

Choose a size that holds your data, your software and your checkpoints. One Python environment with PyTorch takes about 5 GB. One saved container image takes about 6 GB. You can grow the filesystem later, but you cannot shrink it.

These rules apply:

  • One Slurm cluster can use a filesystem at a time.
  • You cannot use a filesystem that belongs to a serverless deployment.
  • Admins of any cluster or VM that uses this filesystem can read the files of all people on it.
Caution:

If the filesystem has a folder with the same name as a new login name, and that folder belongs to someone else, the new person cannot use it. Pick another name. Refer to People and roles.

VPC​

The cluster's VMs connect over a private network, called a VPC. The create page selects your default VPC in the region. You can select another VPC.

If you have no VPC in the region, the create page shows this message:

No VPC in this region. Create one on the VPC page first.

The list does not show VPCs that belong to an InfiniBand cluster, as other VMs cannot join them. Refer to VPC.

SSH keys​

Each person logs in with an SSH key. Refer to SSH to create a key on your computer.

You need the public key of each person. A public key is one line of text that starts with one of these words:

  • ssh-ed25519
  • ssh-rsa
  • ecdsa-sha2-
  • sk-ssh-ed25519@openssh.com
  • sk-ecdsa-sha2-

Keys that you saved in your account appear in a list on the create page. Keys that you add to a cluster apply to that cluster only.

Free compute nodes​

A cluster needs free compute nodes on one InfiniBand network. The create page shows how many are free. Jarvislabs has no queue for compute nodes. If too few are free, try again later or choose fewer compute nodes.

Decisions to make​

Have these answers ready:

QuestionDetails
How many compute nodes?1 to 128. You can add and delete compute nodes later.
What login names?Up to 32 lowercase letters, digits, - or _. You cannot change a login name later.
Who is an admin?The first person is always an admin. An admin can use sudo and read all files on the VMs.
What disk size per compute node?500 GB to 4000 GB. You cannot change this size after you create the cluster. Refer to Storage.

What you need on your computer​

You need an SSH client and scp or rsync to copy files. The examples use a Linux or macOS terminal. You also need to know how to use a Linux shell.