We are planning to buy a large Windows Server 2022 server and use it in a multi-user setting (up to 8 parallel users) to run an open service of bioimage analysis. The server we picked is a dual-CPU AMD 7763 (64 cores each) with 2 TB RAM, 2x A40 GPU and a very fast local scratch storage. The users will basically connect to the server from one of the Windows sessions through RDP.
We have two open questions regarding resource balancing and management:
- How can we limit the usage of the hardware resource per user? In a basic setting: 256 GB RAM, 16 CPU cores and 1/4 GPU cores and GPU memory per user. If GPU sharing is not possible, we could probably have 4 smaller GPUs instead and 1/2 of the sessions without GPU access (if RDP is still possible).
- How can we force that all users from any Windows session share exactly the same software stack and "freeze" all sessions so that for instance the icons on the desktop cannot be altered, no software installed and in general nothing can be modified from the OS partition?
Ideally, any admin software installation / update should also be reflected to all users and we want to avoid local Windows user profiles since all relevant data will exclusively be read and written to the scratch local storage and external storage (mountable partition).