Why is opensm not supported with the WinOF-2 drivers?

Why is opensm not supported with the WinOF-2 drivers?

(cf. https://community.mellanox.com/s/article/differences-between-winof-and-winof-2-windows-drivers)

I would really like to understand the logic and the rationale behind this please.

This means that if you have an externally managed 4X EDR IB switch, and Windows Server 2016, removing the support for opensm renders the ENTIRE configuration/infrastructure null, void, and useless.

I would REALLY want to understand why the decision was made such that ib tools are all not supported in the WinOF-2 driver.

This also means that the ONLY way that the ENTIRE infrastructure can come back online is 1) you have to have at least ONE Linux system with the MLNX linux drivers installed to run opensm and 2) it also means that you just wasted a port JUST for that which means that at the spine level, you can only support 35 Windows Server 2016 systems as opposed to 36.

Why would you do that???

@Chen Hamami​

Thank you.

So…you just basically said the same thing that I’ve already said.

This doesn’t really help answer to question “why” though.

I’m trying to ascertain whether there is fraudulent or false statements made in the marketing materials in regards to the Mellanox ConnectX-4 dual port VPI product since it states that it works in Windows, but the operation is conditional upon either having a managed IB switch, or if you have an externally managed IB switch, that you will need to have a linux system running OpenSM given that OpenSM is only part of the linux driver package and NOT part of the Windows driver package as stated above.

I’m trying to ascertain where undeclared conditional statements about the ability for the product to operate as advertised constitute false and/or fraudulent and/or misleading advertising statements in the marketing and promotional materials published by Mellanox.

Hi Ewen,

The OpenSM code included in WinOF driver is provided as a sample code. The sample code is intended to allow users to test or bring-up the InfiniBand fabric without a management console / switch (to get started). For cluster production environments, Mellanox’s recommendation is to use a Managed Switch / dedicated management appliance (UFM) / a Linux server.

WinOF-2 driver does not include OpenSM sample code and at this point we do not have plans to add SM to the Windows IB stack. In order to establish IB connectivity between Windows machines running WinOF-2 stack, you need to run SM on external entity.

Regards,

Chen

Hi Ewen,

The decision not to implement opensm in WinOF derives from marketing.

Regarding the community article about the differences between WinOF and WinOF-2 - we will fix the statement that opensm is supported in WinOF, as it is not.

Thank you,

Chen

@Chen Hamami​

Thank you, but I am not sure if you understand the problem or the issue with it.

The Mellanox ConnectX-4 dual port VPI product brief (here: http://www.mellanox.com/related-docs/prod_adapter_cards/PB_ConnectX-4_VPI_Card.pdf) states, in the compatibility section, Windows and also that it works with “– Interoperable with InfiniBand or 1/10/25/40/50/100Gb Ethernet switches”.

But that’s not exactly entirely true.

(There is an asterisk next to the supported OS/distribution list, which says: “This section describes hardware features and capabilities. Please refer to the driver and firmware release notes for feature availability.”)

Insodoing, referring to the WinOF-2 release notes (for Windows Server 2016 server), there is no mention anywhere in the release notes that states that the opensm is not packages with the WinOF-2 driver, which means that for someone who is running an unmanaged Mellanox SB-7790 or 7890 externally managed switch, it will give the impression that there is an issue with the hardware when the answer is because the opensm manager is no longer part of the WinOF-2 driver.

In other words, the combination of Mellanox hardware and Mellanox software is insufficient to actually achieve what the product brief and the driver release notes states: that it 4x EDR Infiniband works or is compatible with Windows.

That’s just simply not true.

Nowhere in the marketing or the advertising material, does it state that you need either a MANAGED switch (which has the OpenSM built-in to it) or that you will need another system, running Linux, so that IT can run the OpenSM OR that you will have to download and/or compile the OpenSM from source in order to get it to work/run in a pure Windows environment.

I’m not going to tell you how to run your business, but if it were me, I would be really interested in talking with the Office of General Counsel to make sure that the advertising and marketing material that has been published cannot be interpreted as I have explained it such that it would become a violation of the California Business & Professional Code §§17200 and 17500 and California Civil Code §1770(a) because as far as I can tell, with a set up as I have described, it does not work as advertised/published.

And to be honest and frank, would it really be that difficult to roll the OpenSM back into the next release of the driver? Maybe it’s more involved than I might imagine, being a customer that doesn’t know nor understand the intricacies of doing something like this, but even if it was just a sample code so that we would be able to get the system up and running, all of the advertising and marketing material can be re-written such that if you want something else, you’ll have to do something else instead.

But at least have it in there (like it’s part of the MLNX Linux OFED driver), so I don’t really understand why it can’t be part of the WinOF-2 driver. Yes, I understand that you said that it was a “marketing” decision, but it’s software. The OpenSM driver/module that’s included with the MLNX Linux driver is, if I recall correctly, disabled by default until you enable it.

I’m not a programmer, so I lack the capacity, the expertise, nor the knowledge to be able to understand why this solution, which to an end customer, appears to be something that should be doable, isn’t done, resulting in undue litigation risk exposure to the company, all because of a “marketing” decision which drove a change in software (driver).