Notes from UM Working Group Meeting
 11:00am-1:00pm Thu 6th Sept 2012
 CAWCR BoM Boardroom, 9 East.
 DRAFT  02 by azs, 2/10/2012
 Yix (phone),T.Pugh,  Scott,    Say,   M.Rezny,
 Mjn,        M.Dix,   Lwenming, Ilia,  Joerg,
 Azs.        Apologies: Rab
   Discussion topics (From previous notes): 
   1) UM components progress
   2) TIDG Technical Infrastructure Development Group status
   3) NCI issues
   4) NECTAR Updates
   5) non-ACCESS components such as OPSUI, use of accesscollab
     server etc.
The above was not really followed and we tend to seamlessly
move through the issues.

  i) azs volunteer to keep minutes for this meeting. Will seek
     volunteer for next or future meetings.
 ii) Subset of this group, T.Pugh, Rab, Mike Rezny, Mike Naughton,
     and Martin Dix make up the Technical Coordination Subgroup
     which shall next meet on 20/9/2012 and alternate fortnightly
     with UMWG. next UMWG meeting will be called for early Oct
     ( ?? 11am Thu 4/10/12 ??).
 iii) CSIRO have engaged new staff: Ian Campbell with
      Computer Science background and Python programming
      experience to work on ROSE. It may be beneficial
      to include him in the UMWG.

Unified Model (UM)
1) AMIP run and coupling of vn8.2 is being looked at
   with the aim of making a standard job package available.
   Among the considerations is that there have been significant
   changes to the UKCA package. Martin Dix doing preliminary
2) PS30 NWP package is available at BoM. Not yet looked at.
3) RMIFDF - Project to remove #ifdefs from the UM source
   is progressing well. The first set of patches unifying the
   source for standard and EndGame dynamics have been sent
   to UKMO (Glenn Greed) and he have been able to test it
   with little trouble. Second set of patches have now be
   -- Glenn will be presenting on the RMIFDF project to the
      UM Project Board. It represents major change to the UM
      source so may encounter resistance. Merging with other
      development on top of vn8.2 may not be trivial.
   -- UMWG was not previously aware of the existence of the
      UM Project Board. Being co-UM developers, should we
      have representation?
   -- A write-up on the RMIFDF project have been done and
      updated on UKMO Collab Wiki.
   -- This project represents one of CAWCR contribution
      to the UM collaboration project.
4) Segmentation Fault issue solved by "Bounds Check" compilation
   of offending routine.
   -- This issue have not been fully resolved since the workaround
      using "Bounds Check" compilation work well enough.
       -- Tim suggest the issue may be related to a compiler bug.
       -- M.Rezny commented that the issue need to be documented
          for common benefit, which raised the question that
          currently we have not identified site for these type 
          of information. Mjn sugggest CollabWiki newsgroup.
       -- Ilia plan to investigate the bug further at some stage.
5) As per previous meeting notes, UM8.2 standard jobs are
   available, but further work is required, and will be ongoing
   to further refine and expand what is available.

Technical Infrastructure Work Package.
UM Partner's Board Meeting have taken place.
CAWCR's input to the work package have been sent back to
UKMO and is now awaiting UKMO's response.

ACCESS, UM SUPPORT and related
1) um_help and access_help email addresses are aliases of each
   other and managed using NCI's RT system. "cable_help" currently
   also go to the same list. "nci help" and "cms/climate_help"
   are other help email addresses.
   M.Rezny reminded the group that we are using the RT system
   as guest of NCI. Rita, Margaret and Judy have provided valuable
   help but we need to be good "citizens". We should close tickets
   ourselves which require using the RT web interface.
   Careful usage of email "Subject lines" and removal of
   unnecessary "email tails" is desirable. All correspondence
   are archived by the RT system, "email tails" are generally
   redundant information. Always ensure the correct string
   in the form [ #NNNNN] required for proper
   handling by the RT system is in the subject line. 
   [-> Separate different issues into different subjects]
   M.Rezny will organise training in using the RT system
   some time after 3rd week Oct, when Robin Bowen is around. 
   M.Dix/T.Pugh: Robin has been tasked to develop BoM's
   help/support system for the NCI environment. Quite likely
   it will make use of the available RT system.

NCI Issues
1) NCI have asked us to "slow down" and make proper plans before
   proceeding .
   Ben Evans believe that our requirement for "accessdev" serve
   can be met using the "cloud". 
   T.Pugh stated that (as per NCI request) we need proper
   documentation of our requirements and plans.
   Requirements include:
   i) Server Hardware -- to be met by virtual server in the cloud.
  ii) System Administrator -- NCI and/or BoM/CSIRO/CoE staff
 iii) Operating System / Software stack
  iv) MARS, Mass Store / Archiving requirement.
 Xiao have written up her requirements and sent to relevant
 Request for particular s/w packages, versions etc will
 be addressed later.
2) A meeting have taken place addressing the requirement for MARS
   at NCI. Likely solution is a disk-only MARS utilising openDAP.
   Tan Le is looking at the implementation with time-frame for
   availability targeted for January 2013.
   Of note:  At BoM, 3-4 Tb of data is archived to MARS everyday
   so current capacity of BoM MARS is probably in the Petabytes.
3) The disk /project/access disk on vayu hit the quota limit,
   though quota appeared to have been increased from 2GB to 40GB
   at some stage. There are large amount of CMIP and other
   data which probably can be moved elsewhere. CSIRO staff
   working on CMIP have just returned from holidays and will
   be consulted. [An increase in quota is desirable nevertheless
   for migration work to proceed ]
   There is need to create /project/access/sms and
   /project/access/agreps. We may also want to have
   /project/access/bom for BoM's legacy systems, but
   in general we want common structure for the access
   Should we move /project/access/ to /home/?
   Little interest in pursuing creation of "/access/" link
   to "~access/". Currently "~access/" is set to "/project/access/"
   People who have management permission to "~access/"
   are Martin Dix, Mike Rezny, Margaret Kahn, Scott Wales
   and Say. 
   For solar, BoM have a "Change Management" process in place
   but we probably do not want/need similar arrangement.
   We also do not want a few months gap between request
   and action.
   Quotas for individuals and project will require further
4) Currently BoM access users tend to have ACCESSDIR
   environment variable defined in their ".profile".
   It was suggested that at NCI, we should use "access module"
   instead. Module environment have
   been set up/ being developed for ROSE, cylc and SMS.
   ROSIE is not available and it is not appropriate to
   install it on vayu, nor accesscollab. The new
   accessdev server is most likely the best place for it.
   It was suggested that we need to request/get this
   server ASAP.  Martin Dix, Mike Rezny will address
   this together with Ben Evans (BoM staff ?? ).
   Any request for S/W packages on the server will
   be address later
 5) Shared Repositories
    Access Control List (ACL) management of the repositories
    on access-svn is through trac interface.
    A few people have been designated as "admins". It is believed
    that small group of admins is desirable although backups
    would be required when people go on leave.
    To check if azs is a member of the UM repository admin group
    with permission to update its ACL.
6) UM Tuning.
    Enda's contract is now complete have he have submitted his
    Tim Pugh will forward Enda's report to Group members. Also
    available and will be forwarded are:
     - Paul Selwood's response to Enda's report
     - UKMO IBM P7 report, which Paul Selwood decide to share
       with us in response availability of Enda's report.
       The report address items such as "rework of IBM's vma's"
       "removal of IBM devs", "blocking improvements", OpenMP,
       I/O servers. OpenMP is still an issue, as is hyper-threading.
     - IBM P7 supercomputer is expected to go operational at
       UKMO after the paralympics.
     Further collaboration with Intel in the future is under
     consideration. Collaboration with Fujitsu is taking place
     with a meeting scheduled.
     It is desirable perhaps to combine the collaboration
     with Fujitsu to include Intel, since we will still
     be using intel'l CPUs and compilers.
     Other potential partners are CoE and CSIRO eResearch.
     Workshops to be sponsored by Intel and Fujitsu
 Meeting ends:  1:00pm
 Next Meeting:  11am-12:30pm Thu 4/10/12.


[azs, 2/10/2012] Transfer notes from email 6/9/2012

Last modified 7 years ago Last modified on Oct 10, 2014 2:28:41 PM