The CAE software we develop is using Intel MPI version 4.1 and is typically deployed on clusters with infiniband.
One of our customers is considering the purchase of a cluster that uses omni-path.
Can anyone comment on whether or not I should expect the software to work as-is with omni-path? Or would we need to upgrade our Intel MPI, or perhaps make other code changes?
The Intel MPI Library 4.1 that you're currently running is very old (depending on the update, could be as far back as 2012) and is, in fact, no longer supported. I honestly don't know if it'll run over Intel Omni-Path Architecture (it may, it may not) but you'll most definitely not be getting good performance out of it.
The first official release of Intel MPI Library to support Intel Omni-Path is version 5.1.3. We then made a whole bunch of fixes and tuned the performance in our subsequent 2017 and 2018 releases (we switched versioning schemas mid-way through as well :).
I encourage you to rebuild your application with our latest Intel MPI Library 2018. That should have the latest and greatest support for Intel Omni-Path (and for InfiniBand, and any other fabrics we've continued to tune).
No code changes are required for your code to make it work on Intel Omni-Path.
Hope this helps,
Thank you, that is very helpful. The specific version we have is 4.1.1.036. But given your feedback, we will look into upgrading to the latest version.
Thanks a lot for the quick response.