CDWG Minutes 2013-09-11
- 1 Agenda
- 2 Attendance
- 3 Minutes
- 3.1 Actions items review
- 3.2 Lustre 2.6.0 planning
- 3.3 Lustre 2.5.0 status update
- 3.4 Developer Summit
- 3.5 Patch landing process
- 3.6 LU-2779
- 4 Action Items
- Review action items
- Lustre 2.6.0 planning
- Lustre 2.5.0 status update
- Discuss patch landing process
- Aurelien Degremont (CEA)
- Peter Jones (Intel)
- John Forgan (Xyratex)
- Cory Spitz (Cray)
- Kit Westneat (DNN)
- James Simmons (ORNL)
- Nic Henke (Xyratex)
Actions items review
Xyratex LU-634 update
Nic: The patch attached to LU-634 was just experimental. Xyratex is not interested in the work, no plans to push the patch to gerrit.
Organizations plans for RHEL7 adoption
- LLNL - Maybe Mid-2014. At the earliest.
- Xyratex - not plans for now, sticking with RHEL6.X for a while
- Cray - Only plan to move to RHEL7 when server Lustre support forces them to move
- CEA - No plan before 2015
Cory mentioned that multiple kernel support only supported by ldiskfs, not kernel patching part of Lustre. It would be nice to make that work (or perhaps finally go patchless to make the issue moot).
IU feedback for Intel work on LU-3288 and LU-3490
No "official" response from IU, but Peter reported that at least Stephen Simms seemed happy with the change. Intel and OpenSFS believe the problem is addressed at this time.
Lustre patch submission procedures wiki page
Lustre 2.6.0 planning
- James would like to add checksumming at the lnet layer
- Interop with 2.5.x only; 2.4.x upgrade should be possible
- Testing with Lustre client in 3.11 Linux kernel
- Replace FC18 with FC19 client support (with an eye towards supporting what will become RHEL7)
- UID Mapping and Shared Key (LU-3527)
- IU Primary developers; sharing Lustre over a WAN between different institutions
- Further LFSCK development (LU-1267)
- Funded by OpenSFS; MDT-OST Consistency; MDT-MDT Consistency
Lustre 2.5.0 status update
24 "blocking" issues. Many are test script issues. Newly created HSM tests are many of them.
Tests on one cluster went well with HSM tests, it was only later they realized that a configuration change on another cluster made HSM failures very common.
Possible topics of discussion:
- Client performance
- Patch landing process?
Patch landing process
Some discussion. Newer gerrit 2.6, with its new APIs perhaps enough to provide the desired features of the Xyratex folks.
John suggested that the patches for LU-2779 need attention.
- Make wiki.opensfs.org page describing Lustre patch submission procedures (Chris, everyone)
- Check on status of latest TWG contract. If it has been signed, make sure it is publicized. (Chris)
- Check on LU-2779