Add ocn_glcshelf coupling test #95
Conversation
This PR creates a new SMS_P512_D_Ld5.T62_oEC60to30v3wLI_ais20.MPAS_LISIO_TEST.mpaso-ocn_glcshelf test that runs a G-case with the ocn_glcshelf coupling turned on. That coupling was added in PR #2726, but a test was never added.
|
@xylar , I ran the new test on Chrysalis like: |
|
Great, I'll debug that tomorrow. If it's easy to fix, I will. If it's easier to just comment out |
|
@matthewhoffman, before you move to E3SM, you will need to rename the branch using |
With this fix, effective density smoothing only involves *valid* neighbors of cells.
|
With the fix I just pushed, I'm seeing: Not the fastest test in the world (15 minutes on 8 nodes) so we should improve that with future AIS meshes. |
|
@matthewhoffman, should we do an exact restart test instead of a smoke test? I'm trying the following right now: |
|
ERS also passed: |
There was a problem hiding this comment.
@matthewhoffman, I'll leave it up to you if you want to stick with SMS or switch to ERS.
cime_config/tests.py
Outdated
| @@ -249,6 +249,7 @@ | |||
| "SMS_D_Ld1.T62_oQU240wLI.GMPAS-IAF-PISMF.mpaso-impl_top_drag", | |||
| "SMS_D_Ld1.T62_oQU240.GMPAS-IAF.mpaso-harmonic_mean_drag", | |||
| "SMS_D_Ld1.T62_oQU240.GMPAS-IAF.mpaso-upwind_advection", | |||
| "SMS_P512_D_Ld5.T62_oEC60to30v3wLI_ais20.MPAS_LISIO_TEST.mpaso-ocn_glcshelf", | |||
There was a problem hiding this comment.
@jonbob, should we leave off the P512 and let the infrastructure handle it? I think @matthewhoffman was finding that it ran on 15 Chrysalis nodes by default, which seemed to me like a lot.
There was a problem hiding this comment.
I think we should leave off the P512 and add an entry to the config_pes_tests.xml instead, if necessary. P512 is awkward if it gets run on multiple machines, since the pes/node is variable
There was a problem hiding this comment.
@jonbob , what's the procedure for updating config_pes_tests.xml? Would it be better up update config_pes.xml? I've always been a little uncertain about how to specify compset/grid combinations properly in those files.
|
@xylar , thanks for the quick feedback and the updates! Glad to see you got it running to completion easily. I agree that switching to an ERS test would be more robust, even if it takes longer, but I'll wait to let @jonbob chime in before changing the PR. In terms of timing, the GLC cost is irrelevant, and it's the OCN and ICE cost (as I think we expected with this mesh): My thought is to add this test with the grid currently in the PR (because we need ice-shelf cavities to exist), even if the test is expensive, and then update OCN/ICE to QU120 or QU240 when we have the time to create one (or both) of those with ISC. I will start a PR to update the GLC Antarctica meshes soon, so we can presumably move away from the AIS20km mesh at that time too. What are your thoughts on that? We could reduce the duration from 5 days, so 2 or 3 days if we wanted to make the test run faster and still cover what we want. @jonbob , do you have any advice about this? |
This provides a low res ocean mesh with ice-shelf cavities to permit faster tests of the ocn-glcshelf coupling. Needed mapping files are listed but the filenames are not added yet.
|
@matthewhoffman -- we will probably need a new compset to run this new resolution? The current LISIO compset is hard-wired to use CORE-II forcing: We could add an MPAS_LISIO_JRA1p5 compset to this PR: I tested with the new resolution and this compset and it ran successfully |
|
@jonbob , I actually created a 'GG' compset in my Greenland OCN->GLC PR here: a9224b8
|
|
@matthewhoffman -- thanks for catching that. I think MPASO%DATMFORCED is preferable, and my first shot just a modification of the existing LISIO compset. Apparently it did not get updated when we introduced the DATMFORCED option. I think it's fine having both flavors, though I am not set on the LISIO version -- I mostly was trying to find a compset to test the new resolution and associated files with. |
This commit makes two updates to the new test: * It replaces the MPAS_LISIO_TEST compset with MPAS_LISIO_JRA1p5. The old MPAS_LISIO_TEST will no longer be supported. * It deletes the new test that was added in a previous commit to the e3sm_ocnice_stealth_features suite and instead updates the existing SMS.T62_oQU120_ais20.MPAS_LISIO_TEST in the e3sm_developer suite in a number of ways: 1. uses the updated compset 2. has glcshelf coupling enabled 3. switches from SMS to ERS 4. uses TL319_oQU240wLI_ais20 mesh
|
@jonbob and @xylar , I've updated the test as we discussed this morning. I ran it on chrysalis with: and it appears to have taken just a couple minutes for the ERS test to run and it passes comparisons. But it would be good to have someone else confirm I'm interpreting the test results correctly. Unless either of you see anything amiss, I think this PR is ready to move to the main repo. |
|
closed in favor of E3SM-Project#6437 |
This PR adds a new test that exercises the ocn_glcshelf coupling. That coupling was added in PR E3SM-Project#2726, but a test was never added.