LWG Minutes 2018-04-05

From OpenSFS
Jump to: navigation, search

Attendance

Cray: Ben Evans, Colin Faber
ORNL: James Simmons
HPE: Christopher Voltz, Robert Clark, Olaf Weber
Sandia: Ruth Klundt
SuperMicro: Abe Asraoui
BP: Shawn Hall
Intel: Joe Gmitter, Peter Jones, Andreas Dilger, Amir Shehata

Actions

New Actions Captured:

  • All: Review the circulated roadmap ahead of the next meeting to offer any suggested changes ahead of LUG.

Existing Open Actions:

  • None

Actions Recently Closed:

  • None

Minutes

2.11.0 Release Update
Peter


2.12.0 Release Update
Peter

  • We have started to populate 2.12 page, including starting the feature list: http://wiki.lustre.org/Release_2.12.0
  • On the Intel side, we have a few things planned and are currently assessing FLR plans. We plan to discuss FLR outlook at the LUG developer day.
  • DDN thinks they may have some features ready in this timeframe, but nothing officially confirmed yet.
  • Is there anything on the Cray side to add?
    • Colin: We are still working on a few items and should have a more formal plan in the next few weeks.
  • James, anything from ORNL?
    • James: Finishing up the upstream work will be main focus. It would also be good to target improving lnetselftest. The issue is that it isn’t implemented well and can crash the kernel fairly easily.
      • Andreas: It is good to measure network performance, message latency, etc.. and we do ask people to run it in some scenarios, so it would be worthwhile hardening it.
      • James: We need to keep the same interface, but implement it differently as it was developed under a different set of circumstances.
    • James: LU-10756 - We would like to use the udev events to check on health. Would this be considered a feature?
      • Andreas: Yes, it would be a feature. Some of the guys working on IML would also be interested in this, so it would be good to talk with them about it. They wanted something similar so that IML would be notified when a target starts or shuts down, a client evicted, etc…
      • Peter: As part of your LUG presentation, it would be good to have a list of JIRA tickets needed for prioritization.


Upstream Lustre Client
James

  • The Lustre 2.10 client stuff is pretty much working on ARM.
  • I have been testing with 2.11 servers with and without the 64k patch. Ruth tried with 2.10 servers and had issues. I am wondering if pieces are missing in 2.10 to make it work.
  • I have been talking with Amir about possible back ports. We can roll back a little or move forward a lot for 2.10.
    • Ruth: I set map_on_demand to 16 on the server side and it seemed to work.
    • Peter: People want 2.10 to be as useful as possible, however what people are looking for most on that branch is stability since it is the current LTS stream.


lustre.org
Shawn Hall

  • The group continues looking into Q&A type content and options for moving to move of a Q&A type system.
  • We have run into some hurdles that need discussion.
  • The larger question is how a new system, whether Q/A or forum, how does that integrate with the use of the existing mailing list. If we start using stack overflow as a platform for community questions, how will that work with the existing mailing list?
  • Andreas: If we put it up front on lustre.org it should drive traffic to stack overflow naturally.
  • All agree that stack overflow is more geared to self serve and more beneficial than the mailing list. It comes down to how to manage both the mailing list and stack overflow.
  • Peter: How about those most often answering the questions, such as Andreas and Colin? Is it harder to respond via StackOverflow than just the email or will the new system actually save time?
    • The general consensus is that it should be relatively the same amount of effort to answer.
  • The nature of the stack overflow platform should be an incentive for people to jump in to answer, but we could do things to acknowledge those participating at events like LUG since it is all quantified on that platform.


Other Business

  • Peter: LUG Developer Day
    • Based on who is attending developer day, we have chosen to prioritize those items of interest.
    • We seem to have a full day lined up with good overlap of what people are interested in.
    • Any feedback?
      • James: How long will each presenter get for their talks?
      • Peter: We didn’t put down times, but historically some topics are faster and others go deeper. We should be able to accommodate all the topics in the overall timeframe and we will monitor it as we progress through the day.
  • Roadmap: All should review the circulated roadmap update and offer input ahead of the next meeting.


Next meeting will be on 2018-04-19 at 11:00am Pacific