[openib-general] FW: Open letter to OpenIB membership re-SC05 and response to last board meeting
Ryan, Jim
jim.ryan at intel.com
Mon Oct 17 07:24:14 PDT 2005
I am cross-posting this email to the general mailing list to bring
attention to the activities at SC'05 related to InfiniBand. This is
important work you may not be aware of
Thanks, Jim Ryan
-----Original Message-----
From: Bill Boas [mailto:bboas at llnl.gov]
Sent: Sunday, October 16, 2005 1:21 PM
To: Ryan, Jim; openib-promoters at openib.org
Cc: Rupert Dance; Phamdo, Tuan; tom tucker
Subject: Open letter to OpenIB membership re-SC05 and response to last
board meeting
Jim, and Fellow Members,
I apologize for not participating in the last Board meeting where the
IBTA CIWG and OpenIB presentation was made to those on the call and
voted on.
The slides presented contain serious misconceptions of the state of
the OpenIB core stack (not some of the ULPs) that call into question
the value of the full set of members of OpenIB accepting the proposal
as presented and the subsequent time to market and financial
implications.
First the misconceptions:
Slide 1 Current status - it is rather simplistic to state that each
vendor provides its own unique stack and that interoperability is not
well established.
Many customers like the Labs have mixes of HCA's from one vendor
working, and some in production, with anothers vendors switches. It
is in the subnet managers that the mixing has not taken place so
extensively, and the SMs are involved more in configuring and
management than interoperability.
One of the main goals, already well on its way to achievement, is to
have only ONE CORE stack in the Linux distributions that all IB
vendors, mobo vendors, OEMs (IBM, dell, Sun etc.), database,
filesytem and storage (Oracle, NetApp and DDN etc.) [I'm going to
label these the IB EcoSystem"], AND IB customers (Labs, NSF sites,
enterprise data centers, embedded systems, etc.) ALL use in their OWN
product releases and that customers use at their sites. This
consistency derives from the open source nature of the core stack and
its resilience and readiness for use UBIQUITOUSLY can only derive
from widespread EcoSystem and Customer beta testing.
Slide 2 - Motivation - each of the statements on this slide are
substantially erroneous IMHO.
Interoperability between IB vendor h/w and s/w is regularly deployed
today with what we used to call Gen1 and with some sites already
using Gen2 (i.e. OpenIB release 1.0 candidates) in production end
users i.e customers understand "vendor lock" is probably not a
problem today and will go away with the release of OpenIB Rel 1.0 and
its distribution made available to customers.
Motivation -Cont
The major purpose of the OpenIB sponsorship of the IB infrastructure
at SC05 in November is to demonstrate in a public forum that the IB
EcoSytem and its Customers (abeit in HPC at this time) are "THERE NOW".
I agree that the Enterprise community may not feel that if HPC shows
OpenIB rel 1.0 is "THERE" that more proof maybe needed, but a
Plugfest is not the forum for this, I suggest. The solution is
getting more customers from the Enterprise community to understand
that in an open source ecosytem validation that new or updated
software is ready for their production use is their responsibility
(if they want the full value of open source) or they can turn to a
Tier 1 or 2 or integrator to provide that assurance/comfort for them
at some cost they are willing to bear).
Slide 3 - Technical details -
many details of what needs to be tested and verified in a beta
environment over a sustained period of time are missing (perhaps as
Lamprey and UNH are not part of the ecosystem or substantial
customers they don't yet understand what needs to be done or the open
source and open community environment necessary).
verbs, kdapl, portals NAL, udapl, iser, cm are all missing ..... as
customers we may have issues about the physical layer, links and
cables but the issues are not solvable at a plugfest....
Event Logistics -
SC05 and participation in SCinet05-IB is OPEN to ANYONE, there are
products and members from the whole EcoSystem and Customers (more
than just HPC), there's more variety of hardware, software and
applications there than UNH could ever put together, anyone can
demonstrate or test anything from anywhere in the world because its
open to the Internet, if the exhibitors allow it.
The occurrence of problems will be in the open. And BTW there's data
center fiber, campus, MAN and WAN IB compliance and interoperability
being shown also.
The press will be there and can be shown, or see, IB in action - they
need not be wary of a press release issued by those who have self-
interest in its content.
So here's a suggestion or two for activities that I hope others will
organize at SC05:
1) "IBTA CI and Interop" stakeholders prepare tests to run over
SCinet05-IB (see diagram attached) to ferret out the problems that
they think exist and document the results of those tests on this and
the openib-general mail lists for the whole community to learn about.
If Lamprey and UNH wish to participate and learn I'm sure the IB
EcoSystem will welcome them and help.
2) Most IMPORTANTLY, (as SC05 is not a marketing trade show per se,
but a technical and professional conference where testing and
experimentation is encouraged,) the engineering, software developers,
the QA and support teams of each of the IB vendors, Tier 1and 2s,
Mother board and commodity system vendors, Integrators, software
distributors and Customers (they are all at SC05)
TEST THE HECK OUT the CORE STACK and the ULPs and their customers'
APPLICATIONS (SCinet05-IB should be up and running without
interruption for approx 4 x 24 hour days and open to the world)
3) All members of the IB EcoSystem should also post on these mail
lists and the OpenIB web site any issues with OpenIB, Operating
Systems, and Applications they find on the SCinet05-IB infrastructure.
4) At SC05 OpenIB organizes (Jim can you arrange that?) a meeting of
the membership to review in real time what has been learned and
decide on whether and when OpenIB release 1.0 (and what that actually
is) can be released.
I hope this email generates a lot of discussion and debate......My
hope is that SC05 will prove OpenIB release 1.0 is ready for release
in December 05.. we'll see...
In another email we will update the members on the XNET booth demo an
integrated OpenIB and iWARP stack that Tom Tucker et al. are working
on courtesy of Sandia and NetApp and will be a useful input to the
integration working group referenced in the minutes.
Respectfully,
Bill.
At 04:09 PM 10/12/2005, Ryan, Jim wrote:
>Let me know if comments/corrections on the attached. Note the
>request for the integration team to start work ASAP
>
>Thanks, Jim Ryan, Chairman, OpenIB
>
>
>
>_______________________________________________
>openib-promoters mailing list
>openib-promoters at openib.org
>http://openib.org/mailman/listinfo/openib-promoters
Bill Boas bboas at llnl.gov
ICCD LLNL, B-453, R-2018 Wk: 925-422-4110
7000 East Ave, L-555 Cell: 925-337-2224
Livermore, CA 94551 Pgr: 877-203-2248
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Layout of Infiniband Links in SCinet 05 - IB 10_15_05.ppt
Type: application/octet-stream
Size: 56832 bytes
Desc: Layout of Infiniband Links in SCinet 05 - IB 10_15_05.ppt
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20051017/bc57d2ee/attachment.obj>
More information about the general
mailing list