[Twg] meeting minutes for 2011-03-17

[John Carrier]

In record time, here are the minutes from today's TWG meeting.  We reviewed the draft requirements document I sent to the reflector earlier this morning.  The document includes our recommendations to the OpenSFS board.  

Dave and I ask all TWG members to read the document and provide us with your edits, comments, etc by the close of business this Monday, 3/21.  We will submit the final document to the board on Tuesday.  We want to be certain that we have captured all of the requirements that we discussed over the last couple months and have correctly presented the TWG's priorities.  

If you have different opinions on the requirements or our priorities, then please initiate a discussion of the topic on the TWG reflector.

Thanks for your help. 

--jc
-------------- next part --------------
OpenSFS Technical Working Group
Meeting minutes : 03/17/2011

Concall: start : 9:30a PT, adjourn : 10:43a PT

Next meeting: Thursday, 3/24/2011 @ 9:30a PDT/12:30p EDT
   (Note that the US is on daylight savings time)
   Dial-in numbers are 715-726-4994 or 866-304-8294.  
   Meeting ID and password are 7012090.

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
   Chris Morrone     LLNL           morrone2 at llnl.gov
   Shay Seager       OpenSFS        shay at opensfs.org
   Dave Dillow       ORNL           dillowda at ornl.gov
   Galen Shipman     ORNL           gshipman at ornl.gov
   Sarp Oral         ORNL           oralhs at ornl.gov
   Andreas Dilger    Whamcloud      adilger at whamcloud.com
   Nathan Rutman     Xyratex        nathan_rutman at xyratex.com

Agenda :
   Review the summary document (OpenSFS_TWG_2011-Requirements_DRAFT.pdf)
   that John sent to the reflector  before the meeting started 

Discussion :
   
   John and Dave created the doc from Dave's original notes (see 2/12
   email) and the minutes of our previous meetings.  The document
   includes all of the requirements we have gathered and recommends the
   top priorities we discussed at our 3/10 meeting.  

   TWG members are requested to review the document and provide feedback
   by the close of business Monday, 3/21.  
   
   Note that at the end of the document is a section called "Incomplete
   Requirements," which contains list of topics without descriptions.
   Please review this list and, if an item is one of your requirements,
   please send text to Dave and John with a suggestion of its position
   in the main document.  Any item left in this section will be removed
   in the final version sent to the OpenSFS board.

   Document review :

   - backend storage : Chris asked that we remove the statement
      concerning the commercial viability of ZFS as a backing store.
      LLNL does not believe that distributing ZFS as a binary object
      is incompatible with the GPL.  They know of at least one vendor
      who is planning to support ZFS on a commercial product.
      

   - rates and scales : discuss rows with missing data. We should look
      at the latest 2.0 manual for current Lustre limits.

      max subdirectories - Fujitsu's slides requested 1 billion. There
         was much discussion in our 2/10 meeting on the topic.  There
         really is no limit. Can constrain this by allowing 1 directory
         per core.  
                  2012: 1 million  2014: 10 million
      
      object size limits - currently there is a hard-coded client
         restriction to 2 TBs per object.  It is possible to have the
         server export its object size limit to the client.  Propose to
         scale the object size as 1/2 the OST size:
                                 2012     2014
                  object size   16 TB    64 TB
                  OST size      32 TB   128 TB

      # OSTs vs # OSSs - At our 2/10 meeting, we discussed a 1:1 match
         between OSTs and OSSs. This is never the case in reality.
         Recommend changing the OSTs to be 2x the OSSs and to reduce the
         2014 OSS number to something more probable:
                                 2012           2014
                  # OSS       1 thousand     4 thousand
                  # OST       2 thousand     8 thousand

      
      file size - missing a value for 2014.  4 PB is reasonable, but may
         be better to leave blank if a firm requirement isn't needed.

      metadata rates : stats/s, creates/s etc.  Motivation is improved
         performance of ls -l and du workloads. If no hard requirements
         for stats, then no reason to include them.  Reasonable first
         approximation would be to match the file create rate.

      single process metadata IOPs - these came from Cray's HPCS
         program.  The rates represent requirements for a single client
         and really for 2012, not 2014.
   
   - single file - really is single shared file and requires removal of
      number of OSTs on shared file size. (aka wide striping)


   - varying page-sizes - currently Lustre requires client page size to
      be no larger that server page size.  There are some architectures
      where this could be violated.  This is not a foundational feature
      and will be moved to another section.

   - remove dynamic OST addition/deletion appears to be a duplicate of
      "balancing storage use."  Suggest it be combined with the previous
      requirement.  Also can move adaptive stripe and mutable layouts
      from the incomplete section to this configuration.


   - LNET channel bonding - needs to be moved to peformance section


   - Other comments :
      Need to expand the executive summary at the top of the document to
      explain why the TWG decided to focus on performance rather than
      general administration for its near term priority.

adjourn 10:43 PT