Hello friends,
I was wondering if anyone has had the pleasure of setting up an Intel Omni-Path (now Cornelis Omni-Path Express) network as a fabric in MaaS? Any challenges or pitfalls I should be aware of before I go down this rabbit hole? I’ve searched the intertron up and down and can’t seem to get a solid answer or any cases where this has been done, so any feedback is better than no feedback at all.
Okay, you said “any feedback is better than none”, so here goes.
I don’t have experience with setting up an Intel Omni-Path (now Cornelis Networks Omni-Path) network as a fabric in MAAS. but I have played around with lots of other permutations. Here are some general suggestions based on that experience:
-
Understand Your Network: Before setting up any fabric in MAAS, it’s crucial to have a clear understanding of your network’s architecture and the specific properties and capabilities of the Omni-Path network technology. Decide what you’re trying to do first, with Omni-Path, without MAAS in the loop. Make sure it makes sense and will work. Then bring MAAS into the picture.
-
Documentation: Eat, sleep, and breathe the Omni-Path network doc for the most accurate and up-to-date information. I don’t think that doc is public, unless I just couldn’t find the link. The MAAS documentation also has a dedicated “Networking” subsection under “Explanation”. If you’re at all weak on the basics, please feel free to soak for a while in the networking tutorial on my personal website or some other similar tutorial (lots of good ones out there).
-
Device Drivers: Ensure that you have the correct drivers for the Omni-Path hardware, and that these drivers are compatible with the operating system you’re using with MAAS. Driver compatibility is often a major factor in the successful deployment of hardware.
-
Testing: Before deploying in a production environment, if possible, test the configuration in a non-production environment first to identify potential issues.
-
Community Support: Reach out to the Cornelis Networks communities for advice and potential pitfalls. Warning, you may need to buy support to do this. Other users may have already done what you’re trying to do and can provide useful insights.
Also, see this general warning about HPC that I’ve already offered to someone else. It’s not a simple walk in the park, by any means.
Remember that this is general advice and your experience may vary depending on the specifics of your environment and the hardware in use. If you’re still facing challenges, and you can share more specific details about your setup and the issues you’re encountering – hint, hint: you need to try it and encounter some issues first, maybe? – come back and maybe I can be more useful.
