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This repository has been archived by the owner on Oct 23, 2020.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

Agenda Notes Monday May 1st 2017 11:00 AM

Jump to bottom
Michael Duda edited this page May 1, 2017 · 9 revisions

MPAS-Developer Telecon:

Date: 1 May 2017
Time: 11:00 MDT / 19:00 CET

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

Agenda Topics:

  1. Framework modifications:
  2. io_type not in Streams design document (MH)
  3. Standing topics:
    • MPAS-Atmosphere has renewed interest in generalized variable attributes
    • Subset I/O capability
    • Periodic meshes
  4. Writing history on abort See example for ocean at:

Notes:

Participants: Mark, Adrian, Matt

Framework PRs

  • 1303
    • awaiting feedback from ACME developers to see whether there are any additional changes — if not, then we are ready to merge this ASAP
    • G0 isn’t supported by some compilers; any comments on adding 7 extra characters to width beyond desired precision
    • probably should have this merged within 1-2 days
    • what is threading level about? an e-mail has been sent to Doug to see if he can help us out
      • Per Doug, MPI_Init allows one to request a "threading" level that determines how threads interact with MPI. There are three levels: (1) only one thread is allowed to call an MPI routine at a time; (2) MPI calls can be made in thread-parallel regions, but only one thread will actually do work; and (3) MPI calls can be made in thread-parallel-regions and all threads will do work in the call. In MPAS, we hard-code a requested threading level of (3), but an MPI implementation may return a message indicating that at most threading level (2) is available; this seems to be harmless, in that it just tells a user which threading level is actually going to be used, and it is typically a function of the MPI implementation or system.
    • MPAS-Ocean runs with OpenMPI show erroneously large timer numbers when running with multiple threads
      • we should all be on the lookout for this sort of behavior
  • 1300
    • some code on ‘ocean/develop’ uses interval division and will need changes when merging ‘develop’
    • need to be sure this fixes the “year 1850” problem
  • 1086
    • ready to merge

Documentation for stream option 'io_type'

  • a description of the 'io_type' option was made available in a branch of a fork of the MPAS-Documents repository for the MPAS-Atmsophere guide, but was never pushed upstream
    • create a PR to merge the 'io_type' documentation to the shared Streams chapter in the 'master' branch

Writing streams on model abort

  • Do any other cores write the model state on abort?
    • Yes — the block stream capability can do this.
      • but these aren’t in any strict sense tied to the abort process
    • the landice core checks whether any cores have encountered an error at the end of each timestep
    • a combination of options may be the most useful: let errors propagate up to top-level run routine, and also use block writes when it becomes necessary to do an immediate abort of the simulation
    • block write is useful for very large meshes where large data volumes are undesirable
    • Mark has added pointers in the agenda to lines of code in the ocean core that implement block write capability
Clone this wiki locally