Azure-docs: Non-SRIOV like H16r, CentOS 7.6+ and RDMA = impossible?

[wandhydrant]

Hello! I'm a bit confused about this situation, it would be great to clarify:

_TL;DR: I want to use RDMA/Infiniband on a non-SRIOV Instance like H16r with a recent CentOS (7.8 or if need be 7.7). Am I right this is not possible?_

I'm successfully using CentOS 7.5 but I fear that the necessary kernel drivers (msft-rdma-drivers as present in the repositories of OpenLogic until 7.5) are no longer maintained for later kernel releases.

Currently the documentation says:

On RDMA-capable VMs, add the InfiniBandDriver extension to enable InfiniBand.

But the next sentence clarifies that the InfiniBandDriverLinux is only helpful for SR-IOV:

On Linux, the InfiniBandDriverLinux VM extension installs the Mellanox OFED drivers (on SR-IOV VMs) for RDMA connectivity.

On https://docs.microsoft.com/en-us/azure/virtual-machines/h-series?toc=/azure/virtual-machines/linux/toc.json&bc=/azure/virtual-machines/linux/breadcrumb/toc.json#supported-os-images-linux :

For non-SR-IOV enabled VMs, CentOS-based version 6.5 HPC or a later version, up to 7.5 are suitable. For H-series VMs, versions 7.1 to 7.5 are recommended.

But, as CentOS releases are point-in-time and do not get security updates if you freeze your repositories to an earlier version - I have a hard time seeing "7.1 to 7.5" as recommended. My recommendation in this case would be to only use SR-IOV instances.

Or is it a viable way to update everything but the kernel to the latest CentOS 7? Has this been tested?

---

Document Details

⚠ Do not edit this section. It is required for docs.microsoft.com ➟ GitHub issue linking.

• ID: 5d20f8d6-f126-ea03-1bcd-942fc642edee
• Version Independent ID: 2ffc013c-91a1-73e0-451b-d02a36deb38f
• Content: Azure VM sizes - HPC - Azure Virtual Machines
• Content Source: articles/virtual-machines/sizes-hpc.md
• Service: virtual-machines
• GitHub Login: @vermagit
• Microsoft Alias: amverma

[vermagit]

@wandhydrant Your understanding of the support matrix of the CentOS versions vis-a-vis SR-IOV VM types is correct: <=7.5 for non-SR-IOV types such as H16r, and >=7.6 for SR-IOV types. The IB Linux extension installs only OFED, not the ND drivers for the non-SR-IOV VM sizes.
You make a reasonable point on the need to use newer versions. However, the ND drivers on those, if installable, would be unsupported. I will let others chime in on the feasibility, and instructions for doing so.
Note that the non-SR-IOV VM sizes may be upgraded to SR-IOV in the coming months, so that this bifurcation does not persist.

[wandhydrant]

Thank you very much for this clarification. For me, this issue can be closed.

[mimckitt]

We are currently cleaning up old issues and closing out items that are greater than 90 days old. If an issue is still present, please open a new feedback item on the document so we can prioritize correctly.

please-close