Hello,
We have a single server (2xAMD 128 core, 768 GB, 2x7.68TB NVMe) that we would like to use to host compute jobs from specific users among a limited pool of institutions. We are considering using OpenOnDemand as the interface to access this server. Would this be a good fit for us? All processes would be hosted on this server, including SLURM and authentication, we are not planning to have a separate login server. Is this an issue?
Thanks,
-Jeff
Ultimately it’s a judgement call based off of trade-offs.
The trade-offs as I see them are the time and effort to set OOD up with apps like Jupyter vs. the time and effort to instruct your users on how to run and connect to Jupyter (securely).
To me it makes sense to use OOD to offload that work from your researchers so that they can just do the work they need to do without having to be trained on this or that when they can just click through a web portal instead.
As to hardware, I think you’ll be fine, just given OOD a CPU or two outside of the Slurm allocation. Another issue is memory as folks are using OOD to view & edit files which can utilize some memory in the process. So you’ll just want to carve out a little bit of CPU and memory from Slurm for this.