[openfabrics-ewg] Where do contribute new stuff to OFED scripts?

[Jeff Squyres]

On 9/17/06 9:57 PM, "Michael S. Tsirkin" <mst at mellanox.co.il> wrote:

>>> How about the usual time slot on Monday, 25 Sep?
>>> - Noon-1pm US Eastern
>>> - 9-10am US Pacific
>>> - 7-8pm Israel
>> Hm, might not be simple for me to call in due to a schedule conflict for
>> that week Sep 25-28.
>> However, due to the need of this call, I'll try or at least another
>> colleague will do.
> 
> Why the rush? OFED 1.1 will be out with current methodology.

My point in this whole thread is that there is confusion among the community
-- not so much about *what* we are doing, but *how* we are doing it
(example: taking a snapshot of unstable libiverbs which was then rolled back
late in the release process and then caused other problems -- this should
never have happened in the first place).  Several members of the community
have a different set of ideas in their head about what the process is, and
this is causing both confusion and conflict.

Yes, there may have been an e-mail or two sent about an outline of the
release process somewhere in the past.  But the following is missing from
that method:

- there was no group discussion about the process (the prior e-mails were
mainly information broadcasts), such as what features will be included in
each version
- there is no globally-accessible view of the plan (searching in a web mail
archive is not sufficient) -- something as important as the release plan
should be prominently and obviously available somewhere (web site?)
- there is no tracking of how this plan has changed over time

It's hard enough to ensure that every member of a single, large organization
has all the same information and is fully updated over time.  The problem
becomes significantly more complex when you start involving multiple large
organizations (e.g., OFED).

For example, some of the questions I have asked in this thread have met with
"Well, of course, the answer is X" because it was obvious to some other
member in the group.  But the point was that there was nowhere that I could
go look up the answer.  While I was able to ask and get an answer in this
case, I think there have been several cases in the OFED 1.0 and 1.1 process
where members (including me!) made an assumption rather than asking, and
that assumption later turned out to be different than other members'
assumptions (and causing problems).

I bring up all this because we faced *EXACTLY* the same issue in the Open
MPI group, and we have a lot more communication between the OMPI members
then we do here in OFED (e.g., weekly teleconferences, high-volume
development lists, frequent phone calls an IM between all developers, etc.).
But even with all this communication, and even working with each other for
over 2 years, we still faced exactly the same problems that I described
above (and throughout this thread).  This became a problem in the recent
past with the recent addition of several new members to the OMPI group --
the group became so large that our own communication mechanisms simply did
not scale.  Confusion and some conflict ensued.

We (the OMPI members) finally got together a few months ago and drew up a
draft of simple and lightweight procedures for development and release.
While managing such a large, distributed group of organizations is (and
always will be) difficult, having this pre-release plan agreement has helped
global understanding among all of the members of OMPI about exactly *how* we
get to a specific release.  It has already reduced conflict within the group
because everyone now understands exactly what goes into a release (and what
does not).

*That's* what I want to propose for OFED: a templated (and lightweight) plan
for development and release processes that can then be customized for each
particular OFED series (1.2.x, 1.3.x, etc.).

-- 
Jeff Squyres
Server Virtualization Business Unit
Cisco Systems