CTSM icon indicating copy to clipboard operation
CTSM copied to clipboard
verified publisher icon ESCOMP

Slow single point throughput

Open wwieder opened this issue 3 years ago • 1 comments

Brief summary of bug

  • @adrifoster noticed very slow production in single point cases that was running with PIO_REARRANGER_LND=1 her case directory is here: /glade/u/home/afoster/tutorial_spinup/I2000_CTSM51_spinup
  • @briandobbins suggested using PIO_REARRANGER_LND=2 to try and speed things up.

General bug information

Generally, it would be helpful to figure out what NEON and single point cases need to run efficiently on different systems.

wwieder avatar May 18 '22 22:05 wwieder

This is related to #1743. The fix we did there was just for NEON, but yes it's likely required for all single point cases, and it really should be set in cime.

ekluzek avatar May 18 '22 22:05 ekluzek

Is there anything that can be done to make single point cases run more efficiently? Currently @negin513 can't run 200 years in < 6 hour wall clock time, even with limited history file output. This kind of throughput seems kind of absurdly low. Why have things slowed down so much?

wwieder avatar Sep 30 '22 00:09 wwieder

Yes, we think it's the issue in #1743 where we need a new mesh file for the lightning file.

ekluzek avatar Sep 30 '22 01:09 ekluzek

I'm going to close this since we think the issue had to do with the lightning mesh file.

ekluzek avatar Sep 07 '23 05:09 ekluzek