access_NewSun_005 - ACCESS-NRI/accessdev-Trac-archive GitHub Wiki

#!html
<h1  style="text-align: center; color: green"> CAWCR-BoM ACCESS NWP Ngamai Migration Working Group</h1>

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

Meeting 5: Wednesday 7th August 2013, 9E Meeting Room

Present: Michael Naughton, Chris Tingwell, Robin Bowen, Joan Fernon, Ilia Bermous, Wenming Lu, Zhihong Li, Yi Xiao, Robert Jukic, Joerg Henrichs

Apologies: Asri Sulaiman, Jim Fraser, Ivor Blockley

Welcome Robert Jukic, Ngamai Porting Project Manager
Timeline spreadsheet of major ACCESS-NWP suite porting steps Aug-Nov
Main items in task list
- Item 6: v11 software stack
- Item 17: Building executables
- Item 26: APS1 NWP suites
Other items from Task List
Other business
- NMOC 31-Jul Tech Talk on Ngamai Porting

Spreadsheet ppt file attached below.
Single page identifying month-by-month major steps for CAWCR and NMOC.
Timeline was discussed.
Timings represent best estimate of when work will happen on these steps, realising that delays may occur if problems are encountered in these steps.
More detail on these steps and resource reporting will be developed further.
- ACTION: Robin will draw up draft, discuss with individuals responsible for each of the work items.

Robin reported info from Tim Pugh.
No further steps done on this.
Justin Freeman tasked to work on this, but he's currently on leave for 2 weeks.
Importance of this as a fall-back if there are problems with v12 execs was again emphasised.

Xiao has created ngamai AG1 test NWP suite, based on NMOC operational AG1 suite components, using newly-created v12 executables.
Suite has run successfully through 2 cycles, starting from 21-Jul.
Joan is making some updates to directories, files, scripting of AG1 suite; Xiao will update to this version.
Xiao is planning to start 1-month trial, using suitable recent period, e.g. start from 1-Jul; will check with Joan to get period with matching station-list files for this. Will start running in next couple of days.
Cycle results so far have been examined
- OPS Stats are similar, but don't have good difference measure to quantify differences.
- VAR residuals after first VAR cycle are similar, but differences look a bit larger than expected.
- Need to look at developing ways to estimate OPS differences.
  - ACTION: Chris, Xiao, Zhihong to look at OPS component assessment.

Building executables is broken down into 3 steps (see details)
- source
- software stack
- documented build
v11 sources for most of the required execs have been successfully re-created on solar, checked against existing execs.
v12 execs for ngamai testing have been built for most of the required execs
Documented builds are being worked through:
- running completely from umui is difficult because the umui settings are hard-wired for particular v11 versions of compilers, mpi libraries, gcom libraries; need to identify best way to do the standard v12 builds.
- umui will also be needed on ngamai for ngamai building.
Ilia has reported variations of around 10% in run times with openmpi 1.6.5, up to 30% with 1.6.3.
- Joerg is not seeing variations of this magnitude in his machine acceptance test benchmark runs (N512 vn6.4), which have to be within 5%.
  - ACTION: Joerg to look into variation in Ilia's job timings.
VAR/OPS use build scripts, building does not use varui, opsui; documented builds will be done in terms of these scripts. Same will be true of some of the other non-UM components.
VAR uses source from other svn libraries as well as VAR library; this is additional aspect to be managed in this case.
AG1 UM executable
- Ngamai version including Zhian's physics mods has been built, using extracted source from Xiao, build step from Ilia.

NMOC 31-Jul Tech Talk on Ngamai Porting
- Robin sent email on Monday on this; link is NMOCTechTalks (internal Bureau only).
- For future talks, can NMOC please include ACCESS Ngamai Porting Working Group in the circulation list.