wiki:access/NewSun_008

CAWCR-BoM ACCESS NWP Ngamai Migration Working Group


CAWCR-BoM ACCESS-NWP Ngamai Porting Working Group Meeting Notes

Meeting 8: Wednesday 28th August 2013, 9E Meeting Room
Present: Michael Naughton, Jim Fraser, Chris Tingwell, Robin Bowen, Ilia Bermous, Wenming Lu, Yi Xiao, Joerg Henrichs, Joan Fernon, Martin Dix, Asri Sulaiman(phone)
Apologies: Zhihong Li, Robert Jukic, Ivor Blockley, Peter Steinle

Welcome to Ed Habjan from Oracle attending as observer


Agenda

  • Suites: AG1, AR1, AC1, Disk Mgmt
  • Executables: Build Documentation, Time Variation, Diagnostics
  • Project Management
  • Items from task list
  • Other Business

Suites

AG1

  • DISK usage
    • Details to be discussed between relevant parties
    • ACTION: On-going monitoring
  • Plots
    • Plots now available at close to real time.
    • mjn had look at plot and it lookes comparable to APS1 & AGREPS ensemble members -- OK
  • Gary's diagnostics
    • mjn to discuss with Gary
  • NMOC AG1 suite (Joan)
    • Progressing ... minor problems needed solving
    • Output fields from NMOC AG1 will be archived in MARS7OP
    • MARS retrieval script (mars.sh): NMOC to resolve the scripting
    • Wenming to produce plots when suite trial is going into MARS
    • Verify on ngamai not yet available (can still verify on solar in meantime)

AR1

  • Restarted because pressure level output was missing
  • Have re-run 3 weeks from mid-June
  • Initial verification using laps-verif indicated acceptable results
  • Joerg having problem running Xiao's script for run time-variation investigations

AC1

  • Have run March-April period
    • Delayed by MARS issues
    • Plots have been done which look OK, but no similar data to compare it with
    • Rainval plots to be done
    • MARS archive id has be set to ngaX
    • Switch to current test period from late-June onwards
      • use data from Xiao's ngamai AR1 -- 2-3 weeks of data already available
  • Runtime variation
    • Plus/minus 5% was observed
    • Compared to solar, which had 5-20% variation, fluctuating month to month

ATC1

  • Currently setting up on ngamai; no problems
  • Non-porting issues relating to ACCESS-TC to be discussed in ACCESS-NWP WG

Executables

  • Performance -- runtime variation
    • UKV reruns from a couple of months ago on solar now takng 30-40% longer
    • Joerg still investigating issue with swap_bounds_mv
    • Different MPI library is used on solar
    • Ilia's test config is to do 2 runs within the same job
      • i.e. identical set of nodes are used - test setup suggested by David Singleton
      • On ngamai, second run usually takes considerably longer
        • first runs usually ~ 10% variation
        • second runs more variation, up to 100%
      • Joerg seeing similar variations
      • On vayu, 2nd runs can be faster
      • On solar and raijin - no change
      • Reconfig runs taking between 3.5 to 20 minutes
      • Suggest "sleep" between two jobs to give time for flushing of buffers.
    • Lustre striping being investigated
  • Documented build
    • Changes to UMUI required to include command to use PRG_ENV
    • VAR & OPS uses build procedure outside of VARUI/OPSUI
      • Method to handle VAR/OPS sources to be decided
    • ACTION: azs to continue and follow up with relevant parties regarding build procedures and documentation, and set up initial wiki page

UMUI

  • Updated ghui2.0 to solved bigfont issue implemented on ngamai
    • Some problems remain. azs following up together with Say

PROJECT MANAGEMENT

  • rab providing updates on this WG progress to higher management
  • Further documentation on steps required, rab to email people
  • Some project documentation available through the network: url or s-drive location to be sent, or put on the wiki

/apps

MARS

AOB

  • Robin to report to OSICG
  • Met Office TI collaboration project TAG video-conf last week
  • RDSI setup at NCI -- Repository for datasets of national significance
    • ACCESS NWP, Seasonal, Climate
    • Several Petabytes available in NCI /gdata
    • Historical and future
    • Analyses/forecasts
    • Start thinking of copying data from ngamai to NCI directly rather than going through sam
    • NetCDF format down the road
  • openmp course
  • Resume UMWG meetings to broaden the scope to this meeting beyond ngamai porting to ngamai/raijin/accessdev and other aspects
  • Suspend/resume being worked-on on ngamai
  • MPI library with "THREAD_SERIALIZED" only available with intelMPI on raijin; Ilia investigating




* * * NEXT MEETING: Wed 4th September 2013, 11:00am, 9E Meeting Room



Items arising after meeting for discussions in next Meeting

  • Team members leave plans
  • Items from task list?





--> BACK TO Ngamai Porting main page


[3/9/2013 Edited by azs/mjn ]


Last modified 6 years ago Last modified on Feb 3, 2015 4:33:51 PM