OpenStack¶
OpenStack is meant for custom workloads that do not fit into any other service offered by TU Wien. It gives you access to CPU and GPU resources in the form of simply deployed virtual machines that can then be configured to serve most purposes.
Warning
The OpenStack service is not location-redundant. Hosting mission-critical workloads is not recommended. If you are looking for a virtual machine solution for mission-critical workloads, please check out TUhost.
When should you use OpenStack?¶
The short answer is probably not. We provide access to our compute resources in many other ways that can be used more efficiently. OpenStack is an underlying tool for many of our other services rather than something we recommend for direct consumption. That said, there are a few cases where it makes sense:
- Custom workloads: You have a workload that needs custom-compiled code, direct access, or can somehow not run on a TUhost machine.
- Testing environment: You need a simple, low-power machine to run tests that cannot run in a container or on TUhost.
- GPU workloads: You need a GPU, and our SLURM or Jupyter environments don't fit your use case.
Note
CPU resources are available on demand. All of our GPUs are currently allocated, so GPU flavors are not generally available. If you need persistent GPU capacity, contact us about hosting your own GPUs.
Tip
Kind request: if you intend to create more than 1-2 VMs, please create a private network and use only as many public IP addresses as you absolutely need.
Resources¶
The resources you can request via OpenStack fall into a few categories:
| Resource | Description |
|---|---|
| Compute | CPU flavors, from small 1-vCPU machines to large high-memory or HPC flavors. See flavors. |
| GPU | NVIDIA GPUs through dedicated GPU flavors, currently not generally available. See flavors. |
| Storage | Volumes (Cinder) and shared file systems (Manila/CephFS, e.g. NFS). See storage. |
| Networking | Private networks, routers, floating IPs, load balancers and security groups. See networking. |
Getting started¶
- Log in to OpenStack.
- Create a project or request access to an existing one.
- Create a virtual machine.