[Twg] TWG meeting minutes for 2012-02-23
John Carrier
carrier at cray.com
Fri Feb 24 08:15:12 UTC 2012
Attached are the meeting minutes from this week's TWG call. We spent this meeting reviewing Whamcloud's status of their OpenSFS development projects.
We will have a meeting next week (3/1) to begin our discussion of requirements and roadmaps of new Lustre features.
Dial: 1-715-726-4994, 1-866-304-8294
ID & Password: 72090
Given how quickly time went this week, I would like to propose we plan for next week's meeting to run for 90 minutes.
Please post questions/comments to discuss at lists.opensfs.org<mailto:discuss at lists.opensfs.org>
Thanks,
--jc
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensfs.org/pipermail/twg_lists.opensfs.org/attachments/20120224/6dfba4c1/attachment.html>
-------------- next part --------------
OpenSFS Technical Working Group
Meeting minutes : 2012-02-23
Concall: start : 9:30a PT, adjourn : 10:36a PT
Next meeting: Thursday, 03/01/2011 @ 9:30a PDT/12:30p EDT
Dial-in numbers are 1-715-726-4994 or 1-866-304-8294
Meeting ID and password are 72090
Attending :
Name Organization email
----------------- -------------- ----------------------------
Diego Moreno Bull Diego.Moreno-Lazaro at bull.net
Cory Spitz Cray spitzcor at cray.com
John Carrier Cray carrier at cray.com
Alex Kulyavtsev FNAL aik at fnal.gov
Joshua Walgenbach IU jjw at indiana.edu
Chris Morrone LLNL morrone2 at llnl.gov
Marc Stearman LLNL marc at llnl.gov
Shay Seager OpenSFS shay at opensfs.org
Bob Ciotti NASA Ames ciotti at nas.nasa.gov
Richard Vanderbilt NetApp Richard.Vanderbilt at netapp.com
Andreas Dilger Whamcloud adilger at whamcloud.com
Richard Henwood Whamcloud rhenwood at whamcloud.com
Nathan Rutman Xyratex nathan_rutman at xyratex.com
Agenda :
* Status of Current TWG Projects
- Whamcloud
- IU
- Test bed
* Requirements & Roadmap
- update community requirements
- identify features in need of development
- funneling projects outside OpenSFS into CDWG
* F2F meeting at LUG
- proposal for WGs to meet Weds afternoon
NOTE: Richard Henwood, project manager at Whamcloud, presented the
status of their OpenSFS projects. The discussion of this first
topic consumed most of this week's meeting time. We plan to
continue the agenda starting at "Requirements & Roadmap" next
week.
Status of Whamcloud Projects :
Richard (RH) presented status on the three current projects Whamcloud
is developing for OpenSFS.
= Single MDS performance
- lead engineer : Liang Zhen
- two projects : - parallel directory operations
- SMP node affinity
* parallel directory operations (pdirops)
- code has landed in master
- documentation available in Jira tracking bug (LU-50)
- results show 25% MDS performance improvement for only certain
functions (mknod, open/create)
- pdirops only improves concurrency in the same directory
. when multiple threads access the same directory,
Whamcloud is seeing up to 25% MDS performance gain for
open/close or mknod
. tasks that require multiple RPCs (such as unlink, close)
do not benefit from the new locking mechanism
discussion:
- BobC mentioned that his users do lots of open/closes
and Lustre's long close times are a pain-point. The new
code is bound to highlight new bottlenecks which the TWG
might consider for future development projects.
- The new locking mechanism is currently available only on
the MDS. Andreas said it would be available to the OSS
after the OSD project at LANL (aka Orion) completes.
* SMP node affinity
- Liang is just completing scope statement
- two tracking bugs (LU-55, LU-56)
discussion:
- Cory requested that Whamcloud add document updates to the
tracking bugs. Cray wants to propagate the new calls
through its LND, just as they have with new features
added to ptlrpc.
= Distributed namespace (DNE)
- lead engineer : Wang Di
- project in two phases : - remote directories
- striped directories
* remote directories
- design phase is complete with good feedback from community
- implementation is expected to take 6-7 months with several
intermediate milestones
discussion:
- DNE work is tied to Orion project at LLNL for Sequoia
since Orion restructures the MDS to get rid of Lustre's
VFS hooks and finishes the work to create the OSD
interface. These changes create a cleaner abstraction
between MDTs.
- As a result, OSD must land in master before DNE can.
* striped directories
- the project has not yet started since i depends on the final
remote directories design
= Lustre FSCK
- lead engineer : Yong Fan
- project in 4 phases : - OI scrub and inode iterator
- MDT/OST consistency
- MDT/MDT consistency
- performance
* OI scrub and inode iterator
- this project has been delayed two months while Whamcloud
has reworked the design and recoded the iterator to work
within the OSD framework
- the new design is complete and the project is expected to
complete in March
* MDT/OST consistency
- The delay of this first phase will delay the start of
the MDT/OST consistency check. Whamcloud believes there
is enough time to complete this 2nd phase for the planned
Lustre 2.3 release.
= project risks
- landings
There are a lot of new patches scheduled for the 2.3 release
as part of the OpenSFS and LLNL contracts, as well as patches
not tracked by either project. Whamcloud will be meeting in
Beijing to develop a consolidated landing schedule for 2.3 that
minimizes rebasing and retesting.
- scale testing
OpenSFS is providing a small test cluster at IU for funciontal
testing of DNE. Whamcloud feels that DNE also needs to be
tested at scale with >100 clients. A cluster this size is
currently not part of the plan at IU. Hyperion is not
available because Whamcloud needs access for foreign nationals.
Cray is still trying to put a single cabinet XE at IU, but even
this may not be enough clients.
Status of IU's LustreWAN Project
Dave and John have been working with Steve and Josh at IU to complete
a statement of work for their proposal for UID mapping and alternate
GSSAPI authentication features to facilitate LustreWAN deployments.
discussion:
- We will share the SOW with the TWG after we complete our review
the next draft. The final draft will need approval of the
Board.
- Andreas asked to have the shared key GSSAPI mechanism available
sooner than later. Josh said the security folks at IU would
be adding this feature in parallel to the UID mapping project.
LUG (April 23-25, Austin, TX)
OpenSFS has offered to provide meeting space for its working groups
on Wednesday afternoon (4/25) after LUG adjourns. Both the CDWG and
TWG have expressed interest in meeting. Since we share many of the
WG participants, the meetings will be run sequentially.
Dave and John want to use the TWG time on Wednesday to work on the
requirements and roadmap for new Lustre features.
The OpenBenchmark for Lustre team is refining their charter with the
OpenSFS Board. This working group will meet Sunday afternoon (4/22)
before LUG starts.
If you are interested in attending these OpenSFS sessions, please
plan your travel accordingly.
More information about the Twg
mailing list