[Twg] TWG meeting minutes for 2011-11-14
John Carrier
carrier at cray.com
Thu Nov 17 12:42:43 UTC 2011
Thanks to all who attended the F2F meeting at SC11 this week. (We had a phone line open, but didn't hear anyone join the call.) My minutes from the meeting are attached. Please send additions or corrections to the reflectors.
As discussed at the F2F, there will NOT be a concall this week or next. Attendees at the F2F, in fact, stated their preference to hold concalls only when needed to resolve discussions from the reflectors. Please keep the meeting time in your calendars as a placeholder. If concalls are needed, we will use the Thursday, 9:30am PT slot.
>From the F2F we have some new topics to cover in the next few months:
1) Define the statement of work for IU's WAN project. John and Dave will work with Simms to get a draft together. After TWG review, we will present the SOW to the Board. Please let us know if you are interested in participating in the project advisory committee (PAC) for this project.
2) Update our list of requirements for 2012. Please see the list we created last year (http://goo.gl/cZSWG). This list will drive new development programs and requests for proposals in the spring.
Our next chance for a F2F meeting will be at LUG this April in Austin.
Thanks,
--jc
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensfs.org/pipermail/twg_lists.opensfs.org/attachments/20111117/46191c1a/attachment.html>
-------------- next part --------------
OpenSFS Technical Working Group - F2F @ SC11
Meeting minutes : 11/14/2011
Meeting started 1:40p PT, adjourn : 3:30p PT
Next meeting: TBD
Attending
Name Organization email
----------------- -------------- ----------------------------
Cory Spitz Cray spitzcor at cray.com
John Carrier Cray carrier at cray.com
Steve Simms IU ssimms at indiana.edu
Chris Morrone LLNL morrone2 at llnl.gov
Brian Behlendorf LLNL behlendorf at llnl.gov
Norm Morse OpenSFS normanrmorse at gmail.com
Evan Felix PNNL e at pnnl.gov
Dave Dillow ORNL dillowda at ornl.gov
Galen Shipman ORNL gshipman at ornl.gov
Eric Barton Whamcloud eeb at whamcloud.com
Andreas Dilger Whamcloud adilger at whamcloud.com
Oleg Drokin Whamcloud green at whamcloud. com
Peter Jones Whamcloud pjones at whamcloud.com
Bryon Neitzel Whamcloud bryon at whamcloud.com
Jon Nicholson WTSI jjn at songer.ac.uk
Peter Clapham WTSI pc7 at sanger.ac.uk
Guy Coates WTSI gmyc at sanger.ac.uk
Joe Mervini SNL jamervi at sandia.gov
Nathan Rutman Xyratex nathan_rutman at xyratex.com
Agenda (see OpenSFS_TWG_SC11_final.pdf on the reflector)
* TWG Overview/Status
* OpenSFS PAC update
* OpenSFS Roadmap
* 2012 Development Priorities
TWG Overview/Status --------
TWG made significant progress last year that has significantly
contributed to the stability of Lustre in the marketplace.
PAC Update --------
Dave presented an overview and progress update of the current
development with Whamcloud.
All features have the following milestones:
- Scope Statement
- Solution Architecture
- High Level Design
- Implementation
- Demonstration
- Delivery
[See slides for status of the SMP affinity, DNE, lfsck projects]
The PAC has been posting completed milestone documents to the
reflector. We discussed the need to post this information to the TWG
website or wiki to make it easier for members outside the PAC to
track project progress.
Roadmaps --------
[This was a fairly lively discussion. We jumped around a bit.
Below are my attempts to summarize the main points. Please send
corrections or additions to the reflector. jc]
Contributions :
Using Whamcloud's posted roadmap, John showed that the features
landing through 2.4 are all community contributions. We next need
to see what contributions from other vendors are in progress that
we can place on this chart.
Eric said that Whamcloud is being very protective about what could
land because of their concerns for stability. Their defensiveness
should not be construed as competitive self-interest. Instead,
they know that there are many features in the works, but they
don't show up on the roadmap unless there is an actual scope or
effort estimate associated with the feature.
For example, NRS is in progress, but is waiting testing on large
scale resource. Someone asked if it could be included in the code
without being turned on to make testing of the feature easier.
Andreas said this was the way size on MDS was supposed to be
evaluated. SOM was there, but not really tested. NRS is
definitely in the datapath, but review cycles have been long.
OpenSFS needs to provide a mechanism for collecting information
from contributors. We need to follow the pattern
Whamcloud/OpenSFS are using with the PAC to define milestones,
which should be completed overtime, not when the implementation is
done. Recall, Whamcloud asked a year ago that new code be
delivered in small, easily inspected chunks.
We agreed that this is where a wiki and a submission template
could help. It would help contributors formulate their
submissions, generate a delivery caldendar, and provide a target
for unit test results. Items to track on the wiki for each
submission are : exec summary, sponsor, size, expected completion
dates, risk, stability ranking, documentation, test results.
Whamcloud already has something similar in place. Eric said Bryon
can help organize.
Future releases :
We came back to the issue of how to track what new features have
been proposed. The Whamcloud "community roadmap" only reflects
those features with resources estimated and assigned. Whamcloud
has a parallel list of features not yet committed. OpenSFS needs
to extend that to include features being developed by other
contributors.
Eric noted that for a 2.2 release in March, we would need a freeze
in December. 2.3 would then be frozen in June. Andreas(?)
suggested that we ask contributors of new features beyond the 2.3
release to present their proposals at LUG.
Test hardware :
We discussed what resources are available for testing. For
example, NSF is sponsoring Probe--could that be used? Storage
would be local to the cluster node. Andreas said that experience
at Sun was that the 90day burn-in period was not sufficient. It
is clear that we need more test equipment. The blocker for
OpenSFS is that our contractor (Whamcloud) has foreign nationals
that need access. This puts resources like Hyperion out of reach.
ORNL or IU can host donated equipment. We need storage and client
clusters. John to investigate whether Cray can donate a cabinet
of XT to provide 96 client nodes. IU and ORNL have the experience
with Cray to operate the system at their sites. Dave noted that
the contract with OpenSFS requested 4 MDSs and 4 OSSs for testing
of DNE. So we still need servers and storage.
Conference calls :
The preference is to continue to have discussions on email as much
as possible. Everyone agreed that the call should only be used
when email to the reflectors are insufficient. Past meetings for
the discussion of NRS were sited as an example of the value of the
calls. Otherwise, the consensus was that the calls should only be
organized when needed. TWG members should keep the spot open on
their calendars, but we don't need the weekly meeting unless the
co-chairs call the meeting.
Mailing lists :
We currently have multiple "discuss" lists. Why? Whamcloud
started theirs after noting some posts to lustre.org were getting
delayed. lustre.org, however, is still operating. We agreed
that twg at lists.opensfs.org and discuss at lists.opensfs.org should
still be used.
The lustre.org email reflectors serve a purpose since they domain
is well known. lustre-discuss will be difficult to transition but
lustre-devel is such a small community that someone just needs to
say the discussion has moved to lustre-devel at lists.opensfs.org and
it would. [Andreas has already started including the opensfs.org
list in his replies to lustre.org email.]
Chris suggested that we create the site lustre.opensfs.org to
reference opensfs.org. We could give this high google search
rankings if all OpenSFS members put a link on their websites to
it. [note: lustre.opensfs.org does not exist]
2012 development priorities --------
We discussed the need to update our list of community requirements,
which we started at SC10 (http://goo.gl/cZSWG) and submitted to the
OpenSFS to recommend the current metadata projects. We will start
discussions on the TWG reflectors, but need to encourage non-members
(eg, EOFS) to include their suggestions as we did last year.
TWG members should start the discussion on the email reflectors. We
can hold concalls after the first of the year, as needed, to organize
the information. Please review last year's list and make new
suggestions to the OpenSFS email lists.
OpenSFS has money available for new development in 2012. When the new
list is complete, we will need to submit our recommendations to the
Board. Norm said that OpenSFS is a private corporation and does not
have to follow the formal government procurement procedures as we did
in 2011. This leaves us open to consider issuing a call for proposals
as well as issuing a formal RFP. This could encourage smaller
programs, such as IU's WAN proposal, to come forward.
The important point is to get the discussion of requirements started
again on the reflectors. We should be looking at short and long
range projects, which can include foundational features to prepare
Lustre for exascale.
Other topics :
IU's WAN project :
The Board approved IU's WAN proposal. The TWG needs to develop
the statement of work (SOW) for the Board. At the end of the
meeting, as people were leaving, Dave and John talked to Simms
about getting a first draft started before the first of the year.
As part of the SOW, we expect the Board to assign a technical
representative to oversee the project on behalf of OpenSFS and to
create a project advisory committee (PAC) from the TWG membership
to assist this person. This is the same process we are using for
the Whamcloud development project.
Discussion of the SOW will be kept inside OpenSFS on the
twg at lists.opensfs.org reflector only.
Meeting adjourned at 3:30 PT
More information about the Twg
mailing list