[Ofa_boardplus] Jason's draft preso to the Linux Plumbers' Conference

[Jason Gunthorpe]

On Thu, Aug 31, 2017 at 08:20:35PM +0000, Woodruff, Robert J wrote:

> >OFA *should* be using this resource to test upstream linux around
> >the rc candidate series so we can have *tested* RDMA in all major
> >upstream linux releases. This gives the best >chance that a user
> >will have access to a tested RDMA stack from their distro.
> 
> I also think that the hardware interop testing activity is also very
> important and valuable and would not recommend it be replaced by an
> upstream testing activity but instead added as an additional testing
> effort.

This is why I said the program should not bundle so much under a
single logo. Make the software testing distinct and independent of the
hardware testing.

I looked at the lasted hardware list from the last plug fest,
and honestly, it is pretty thin and largely obsolete. I don't know
what value you are seeing here.

I can't imagine any purchaser or integrator getting anything useful
from this list. Have you looked at it? OFA just spent a bunch of money
to test a NE020. This is an obsolete, unbuyable card with a known
broken kernel driver. It is totally absurd that any time at all was
spent on this.  The other hardware isn't much better.

As a sort of outside observer, it appears to me the hardware
vendor member companies do not see enough value in the activity to
even donate latest hardware revisions, so I deeply, deeply, question
your assertion that it is 'very important and valuable'.

.. and the more I look at this the more baffled I become at the idea
this activity is the foundation of the OFA. Sorry Jim :(

> The question comes down to finding a way to fund the
> additional upstream testing at UNH-IOL, which is not self-funded
> like the logo program is. If the OFA wanted to directly fund some
> activities, then funding the testing of the upstream code during the
> RCs of the kernel and rdma-core seems like it would be a very good
> investment.

Yes, I agree.

Split the logos. Allow the IWG participants to decide what activity
they wish to fund. If Intel wishes to continue to fund hardware
testing of an obsolete and unbuyable HCA, then fine by me...

I suspect the reason this keeps going (aside from inertia) is because
there is some value in the software testing. Maybe that isn't even
true, I'd be interested to hear Doug's take on what the software test
suite accomplishes.

Jason