Getting Started

Quick start guide on accessing PRESCIENT

Prerequisites

You will need:

• An identity (id: name.# for OSU) or a sponsored guest account.
• A laptop or workstation you can install a VPN client on (macOS, Linux, Windows, or iOS/Android).

Step 1 — Request an account

TODO: describe the actual request mechanism once it is in place

Step 2 — Set your password

Once your account exists:

1. Visit the testbed account portal at https://auth.resnull.party
2. Log-in using your one-time password. You will also be asked to set up a TOTP authenticator for multi-factor authentication (MFA). You can use Duo for this.
3. After logging in for the first time, go to settings and change your password as this one-time password is marked expired and you will be prompted to change it if you try accessing any of the testbed servers anyway.

TODO: paste the account portal URL and a screenshot of where to add the key.

Step 3 — Enroll your device in the VPN

You join the PRESCIENT VPN (a private WireGuard mesh) using your id and MFA.

1. Install the Tailscale client for your OS: https://tailscale.com/download

2. Point it at the PRESCIENT control server:

   https://vpn.resnull.party
   

   Using the tailscale CLI:

   tailscale up --login-server=https://vpn.resnull.party --accept-routes
   

   In case you are switching from another account in tailscale:

   tailscale up --login-server=https://vpn.resnull.party --force-reauth --reset --accept-routes
   

3. Your browser opens to the SSO login. Sign in with your id and complete MFA. This step may be automatic if you recently authenticated through the SSO portal.

4. Approve the device. Once enrolled, you can reach *.prescient.internal hostnames directly.

TODO: confirm the final --login-server URL and document the iOS/Android flow (which uses a ?key= URL instead of the command above).

Verify VPN connectivity

ping <host>.prescient.internal

If this fails, see Help & FAQ.

Step 4 — First SSH

ssh id@<host>.prescient.internal

User your SSO password here and you should see the cluster MOTD after log in. If you cannot SSH:

• Confirm you are connected to the VPN (tailscale status).
• Confirm you have been granted the right access — without it, SSH is rejected at the PAM layer.

Step 5 — First GPU job

Once on a compute host:

# Interactive shell with one GPU for 30 minutes
srun --gres=gpu:1 --time=0:30:00 --pty bash

# Inside the allocation, you should see exactly one GPU:
nvidia-smi

When you exit the shell, the allocation is released.

See GPU Compute for the full SLURM workflow, container usage, and storage layout.

Where to put files

A two-minute summary; see GPU Compute → Storage for the full version.

Path	Use it for	Persistent?	Shared across hosts?
/home/$USER	code, configs, important results	✓	✓ (via NFS)
/cache/users/$USER	venvs, working datasets, model outputs	✓	✗ (per host)
/scratch/jobs/$SLURM_JOB_ID	intermediate files within a job	✗ (deleted at job end)	✗
/shared/hf-cache	HuggingFace model cache (community)	✓	✓

Next steps

• GPU Compute — SLURM, containers, storage, common workflows.
• Help & FAQ — troubleshooting and who to contact.