[Ofa_boardplus] Voteable item for Thursday's Board Meeting

Doug Ledford dledford at redhat.com
Thu Mar 15 09:56:45 PDT 2018


On Thu, 2018-03-15 at 16:00 +0000, Bernard Metzler wrote:
> Hi Doug,
> 
> Many thanks for the detailed reply. Much appreciated.
> 
> Sorry to hear about your trouble to format the reply as intended
> - so please let me switch to plain ASCII, inserting comments
> accordingly (I prefer that anyway). I deliberately prepend another
> '>' to my original text to fake ASCII history.

Thanks ;-)

> -----Doug Ledford <dledford at redhat.com> wrote: -----
> 
> > To: Bernard Metzler <BMT at zurich.ibm.com>, Paul Grun <grun at cray.com>
> > From: Doug Ledford <dledford at redhat.com>
> > Date: 03/14/2018 06:11PM
> > Cc: "ofa_boardplus at lists.openfabrics.org,"
> > <ofa_boardplus at lists.openfabrics.org>
> > Subject: Re: [Ofa_boardplus] Voteable item for Thursday's Board
> > Meeting
> > 
> > I'm sure Paul will probably write something more meaningful than I
> > will here, but my answers are inline below:
> > 
> > On Wed, 2018-03-14 at 17:27 +0100, Bernard Metzler wrote:All,
> > > I have to admit I did not actively contribute to the mission
> > > statement discussion for some time. But let me share with you what I
> > > think when I just read the proposed new mission statement.
> > > 
> > > It would reflect quite a big change of the mission of the OFA. To me,
> > > old keywords were
> > > open source software development, test, licensing, support
> > > vendor independence
> > > unified, cross-platform, transport-independent software stack for
> > > RDMA
> > > standards-based software
> > > 
> > > 
> > > So, to my understanding, the current mission of the OFA is to develop
> > > and maintain an ecosystem to ease and promote the usage of RDMA
> > > fabrics, while closely following given standards.
> > 
> > I would argue that that mission has been met.  To stick with that
> > mission now has little to no meaning, aside from maintaining the
> > already written software stack.  But, regardless of the old mission
> > statement, the OFA has done more, and will continue to do more, than
> > just what this embodies.  Hence, the need for a new mission statement
> > and the bylaw rewrite.
> > 
> 
> I came into OFA late, in a phase where the RDMA ecosystem
> was already maturing, I don't know much about the early years.
> Maybe, OFA now has just fulfilled its original mission and can
> declare victory? That software stack now lives in Linux, and it would
> likely stay there if OFA would no longer exist.

It would, and that was one of the options on the table I think.  The
question presented itself earlier this year: do we, the OFA, continue
with some new understanding of purpose, or just declare victory, fold
up, and go home.  I certainly thought there was more to be done in the
realm of fabrics.  While the stack is in the linux kernel and a separate
couple of user space projects, there is lots more that could be done in
terms of making it well known to people outside of the HPC space, to
bringing new features and enhancements to the existing stacks, etc.  So
I don't think going home is a good idea.  I think there is still more to
be done ;-)

>  In
> that respect, the scope of OFA may change to an influential
> role.
> But, in any case, I understand that there is a role change.
> 
> > > 
> > > Here are my main concerns with the proposed new mission statement:
> > > It  completely lacks the term 'open source software', which was
> > > essential for the success of OFA so far.
> > > 
> > 
> > No one disagrees that open source is certainly the preferred means of
> > writing software (although it wasn't the only one the OFA
> > used...after all the early RDMA stack was dual licensed GPL/BSD just
> > so it could be used in proprietary software, and for a long time OFED
> 
> Right, it was open source with a focus on widespread usability.
> It was meant to be as minimum restrictive to deploy as possible. We shall
> stick to that policy and make it explicit. The mission statement shall 
> reflect common agreements, and not assume them implicitly.

Ok.  I'm not opposed to putting open source in the statement, it just
has to be word smithed in.

> > shipped proprietary bits).  But the mission statement is trying to be
> > more general than just the software OFA members produce.  It's trying
> > to state that the OFA is trying to advance both the hardware the
> > stack runs on (be that IB, OPA, standard IP with DCB extensions, or
> > just TCP) and the software.  Part of getting the technology more
> 
> How can we want to advance hardware - as an alliance including
> vendors of different, and potentially even competing hardware.
> That sounds tricky. Our role here might better be to get all vendors
> on the table to stay committed to certain interfaces, and evolve
> them together, to meet common needs. IMHO, this alliance does not
> need to become more hardware focused, but more application/user
> focused.

I don't disagree with this.  The advancement of the fabric hardware is
through the evolution of the existing and/or new software APIs that the
OFA is interested in.  Right now that's the kernel level RDMA core,
drivers, and ULPs and the user space rdma-core and libfabric stacks.  So
let's say a new bit of hardware comes along that is RoCEv3 and the
stacks would need to be extended to support it.  That would be in line
with the new mission statement, and that does indeed advance that
hardware.  There is no attempt to play favorites, just an attempt to
keep the upstream stacks up to date with evolving hardware trends.  If
supporting some new hardware required an entire new stack of some sort,
then I think it be a discussion item whether or not the OFA undertook
the process of incubating a stack for it.

But in the end, even as a vendor with no hardware stake at all, I still
have a vested interested in seeing the hardware uptake increase.  My
desire for this is born of the fact that as the hardware uptake goes up,
that means the hardware is more ubiquitous and people are doing more
things with it, which means the selection of RDMA enabled apps will be
going up.  And, as you point out, we need to be user/app focused.  To
me, they go hand in hand.  When you promote one, you are promoting the
other.  I don't want the OFA promoting specific hardware, I want the OFA
promoting RDMA based programming techniques and APIs and bringing them
to more user apps, or bringing more kernel ULPs that users can leverage.
 Doing so, IMO, helps drive the hardware uptake and was the reason why I
thought the statement should be broader than just the software so that
it would catch this nuance of intent.  But that could also be done by
leaving the statement broad at the "what do we do" point, and then
focusing down more accurately at the "how do we do it" point.

[ I snipped the rest only because it's about time for the meeting to
start anyway ]

-- 
Doug Ledford <dledford at redhat.com>
    GPG KeyID: B826A3330E572FDD
    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.openfabrics.org/pipermail/ofa_boardplus/attachments/20180315/dfd5e4bc/attachment.sig>


More information about the Ofa_boardplus mailing list