ESCOMP / POP2-CESM

Parallel Ocean Program (POP2) in CESM
http://www.cesm.ucar.edu/models/cesm2/ocean/
4 stars 24 forks source link

Add 120 task (and maybe 160 task) decomposition for gx3v7 #66

Open mnlevy1981 opened 2 years ago

mnlevy1981 commented 2 years ago

Description of the issue:

The UCI supercomputer has 40 cores per node, and I set up an 80 task decomposition to spread POP across 2 nodes when running gx3v7; they have tried running with 120 and 160 nodes, and run into some issues (see bottom of this section). The auto-decomp tool doesn't provide a very good distribution for those task counts at this resolution, so I'd like to add

<decomp nproc="120" res="gx3v7" >
  <maxblocks >1</maxblocks>
  <bsize_x   >10</bsize_x>
  <bsize_y   >10</bsize_y>
  <nx_blocks >10</nx_blocks>
  <ny_blocks >12</ny_blocks>
  <decomptype>cartesian</decomptype>
</decomp>

<decomp nproc="160" res="gx3v7" >
  <maxblocks >1</maxblocks>
  <bsize_x   >10</bsize_x>
  <bsize_y   >8</bsize_y>
  <nx_blocks >10</nx_blocks>
  <ny_blocks >16</ny_blocks>
  <decomptype>cartesian</decomptype>
</decomp>

to bld/generate_pop_decomp.xml at some point.

The "issues" I've alluded to only seem to be present on their machine, which is using intel 2018.0.3 - there is a crash in running_means_mod.F90 when ladjust_bury_coeff = .true., but I can't reproduce it on any other machine I have access to. I'm hopeful that it is a compiler bug and updating to a more recent compiler will make it go away, but if it persists I'll open a new issue ticket (I have some ideas on how to investigate if I need to).

Version:

Machine/Environment Description:

GreenPlanet (UCI super computer)

Any xml/namelist changes or SourceMods:

n/a