Skip to content
This repository has been archived by the owner on Oct 23, 2020. It is now read-only.

Agenda Notes Monday January 23rd 2017 11:00 AM

Jump to bottom
Michael Duda edited this page Jan 23, 2017 · 4 revisions

MPAS-Developer Telecon:

Date: 23 January 2017
Time: 11:00 MST / 19:00 CET

Call-in number: 1-866-740-1260
Access Code: 4978161

Agenda Topics:

One of the foci of this telecon concerns progress on several infrastructure-related items that have been discussed in the past:

  1. Timekeeping issue affecting ACME runs starting in year 1850 has been narrowed down to mpas_calibrate_alarms routine
  2. PIO performance issues with forcing
  3. Integer overflow in ESMF time interval code - See branch https://github.com/mgduda/MPAS/tree/framework/esmf_i8_intervals
  4. Framework modifications:
  5. Separate opening of streams from mpas_dmpar_init #1194 - need to work on a solution
  6. Design Document Discussions

Notes:

Participants: Xylar, Matt, Mark, Michael

Not much to say about items 1, 2, 3

Mark is satisfied that the move of COMPASS in #1160 will work, looks ready to merge

  • Xylar: is there anything hard-coded in workflows that depends on the path to COMPASS?
  • Mark: probably not, since we first cd to proper directory
  • Matt will merge in the next few days

For PR #1192, may implement this using a general function for expanding filename templates

PR #1201 is meant to replace #1119

  • ACME developer proposed #1119 as a fix, but this breaks block streams (but these weren’t in version ACME developer was working with)
  • Matt has sent #1201 to ACME for testing
  • For now, just get initial feedback from other MPAS developers
  • just move call to open_streams to mpas_framework_init_phase1 right after call to mpas_dmpar_init()
  • open_streams only called when running stand-alone MPAS, but in ADME, we get a communicator and therefore didn’t call open_streams, and so all MPAS printout went to ACME streams, and all tasks were writing even for optimized builds
  • also issue to send output to /dev/hull
  • for now, this allows excessive output to go to /dev/null (and not worry now about problems inherent in that)
  • Matt: will report back when he hears back from ACME, update status

Logging streams module

  • handful of issues when redirecting output to /dev/null
  • can’t have two units open to same file: problem when running ice and ocean in same execution
  • scratch files can solve this issue, but has performance issues
  • each write to stderrUnit or stdoutUnit would be replaced with some new logging function
  • maybe have stable API, then implement in phases
  • homework: read design document and provide comments
  • close requirements by Thurs

Mark: heads-up PR #1151

  • will separate into two PRs, one for core-specific, one for framework
  • need to test whether framework changes work for MPAS-Atmosphere
  • some of the changes are not bit-for-bit (ocean only?) because of order of operations
  • potentially four PRs: {bit-for-bit, non-bit-for-bit} x {core, framework}

Probably need a v5.1 bugfix release in a week or two to address several issues in MPAS-Atmosphere v5.0

Next telecon: Tuesday, 31 January

Clone this wiki locally