1. What Jetstream2 is
Jetstream2 is a research cloud funded by the NSF and run by Indiana University. You create Linux virtual machines (Jetstream2 calls them instances) in a web browser and get full administrator rights (sudo). A VM can run for an hour or for months.
What you get on each VM, as tested by Research Computing:
- Ubuntu 24.04 (or Rocky Linux) with Docker and Python already installed;
- a web shell and a web desktop in the browser, plus SSH;
- the Jetstream2 software collection (R, RStudio, MATLAB, Intel compilers, Miniforge, Apptainer and NVIDIA HPC SDK) through
module load; - on GPU sizes, an NVIDIA A100 (part of a card, or a whole card) with the driver and CUDA already set up.
2. When to use it
| You want to | Use |
|---|---|
| Run batch jobs, Slurm or MPI on campus | HPCC |
| Run containers or GPU jobs on Kubernetes | NRP Nautilus |
Have a Linux machine you fully control, with sudo |
Jetstream2 |
| Log in to a GPU and work on it interactively | Jetstream2 GPU sizes |
| Host a web app, database, Jupyter or RStudio server | Jetstream2 (or Nautilus for containers) |
| Give a class identical VMs | Jetstream2 |
| Work with protected data | Not Jetstream2: see secure research computing |
3. Sizes and service units
Time on Jetstream2 is counted in service units (SUs). A VM uses SUs for every hour it exists in a running state, at a rate set by its size (its flavor). A shelved VM uses none.
Common sizes (from the Jetstream2 flavor list):
| Flavor | Cores | RAM | Disk | GPU | SUs per hour |
|---|---|---|---|---|---|
| m3.tiny | 1 | 3 GB | 20 GB | none | 1 |
| m3.small | 2 | 6 GB | 20 GB | none | 2 |
| m3.quad | 4 | 15 GB | 20 GB | none | 4 |
| m3.medium | 8 | 30 GB | 60 GB | none | 8 |
| m3.large | 16 | 60 GB | 60 GB | none | 16 |
| g3.medium | 8 | 30 GB | 60 GB | 25% of an A100 (10 GB GPU memory) | 16 |
| g3.large | 16 | 60 GB | 60 GB | 50% of an A100 (20 GB) | 32 |
| g3.xl | 32 | 120 GB | 60 GB | a whole A100 (40 GB) | 64 |
Larger CPU sizes, large-memory sizes, and L40S and H100 GPU sizes also exist. Some need a request to the Jetstream2 help desk.
Some arithmetic to plan with:
- an m3.quad running every working day, 8 hours a day, for a month uses about 700 SUs (4 x 8 x 22);
- a g3.medium left on for a week uses about 2,700 SUs (16 x 24 x 7);
- shelve VMs you are not using. Shelving is the single biggest saving.
4. Three separate resources
Jetstream2 CPU, Jetstream2 GPU and Jetstream2 Large Memory are separate ACCESS resources. When you exchange ACCESS credits, pick each one you need. Having CPU time does not let you start a GPU VM. One ACCESS credit buys one Jetstream2 SU.
5. The path to a running VM
- Get an ACCESS account and project. An Explore project is usually enough to start and is often approved within a few business days. Follow KB009: Using NSF ACCESS.
- Exchange credits for Jetstream2 (CPU), and Jetstream2 GPU if you need GPUs, in the ACCESS allocations portal.
- Sign in to Exosphere at jetstream2.exosphere.app with your ACCESS identity.
- Create a VM, then connect to it: Your first Jetstream2 VM.
- For GPUs, see GPU VMs on Jetstream2.
- To manage your VMs and your SUs, see Managing Jetstream2 VMs and SUs.
6. Your responsibilities
- You are the administrator of your VMs: install updates, use SSH keys, and open only the ports you need.
- Keep a copy of important data somewhere else (for example CephRDS or an OSN bucket). VMs and allocations end.
- Follow the Jetstream2 acceptable use policy and your ACCESS project's terms.
7. Help
- Jetstream2 documentation: docs.jetstream-cloud.org
- Jetstream2 help desk: help@jetstream-cloud.org (VM problems, larger sizes, GPU access)
- ACCESS questions: ACCESS support
- Choosing a platform, or an ACCESS request: contact Research Computing