File size reduction #9
Conversation
jjguerrette
requested review from
jjguerrette,
byoung-joo,
liujake and
svahl991
| <var_array name="scalars"/> | ||
| <var name="initial_time"/> | ||
| <var name="xtime"/> | ||
| <var name="cldfrac"/> | ||
| <var name="re_cloud" packages="mp_thompson_in;mp_wsm6_in"/> | ||
| <var name="re_ice" packages="mp_thompson_in;mp_wsm6_in"/> | ||
| <var name="re_snow" packages="mp_thompson_in;mp_wsm6_in"/> | ||
| <var name="u"/> | ||
| <var name="w"/> | ||
| <var name="pressure_p"/> | ||
| <var name="pressure_base"/> | ||
| <var name="rho"/> | ||
| <var name="rho_base"/> | ||
| <var name="theta"/> | ||
| <var name="theta_base"/> | ||
| <var name="relhum"/> | ||
| <var name="uReconstructZonal"/> | ||
| <var name="uReconstructMeridional"/> | ||
| <var name="surface_pressure"/> |
There was a problem hiding this comment.
Looks like there is one column misalignment, which leads to spurious diffs. Need to re-align those lines.
| call MPAS_stream_mgr_read(domain % streamManager, streamID='static', whence=MPAS_STREAM_NEAREST, ierr=ierr) | ||
| if (ierr /= MPAS_STREAM_MGR_NOERR) then | ||
| call mpas_log_write('********************************************************************************', messageType=MPAS_LOG_ERR) | ||
| call mpas_log_write('Error reading static fields', messageType=MPAS_LOG_ERR) | ||
| call mpas_log_write('********************************************************************************', messageType=MPAS_LOG_CRIT) | ||
| end if | ||
| call MPAS_stream_mgr_reset_alarms(domain % streamManager, streamID='static', direction=MPAS_STREAM_INPUT, ierr=ierr) |
There was a problem hiding this comment.
I think we should make this static stream read optional for now (e.g., through an extra namelist parameter 'separate_init_stream' or something more meaningful?), single file restart/init initialization capability should remain unchanged. It may be more acceptable for the model group if we want to commit this change to the official MPAS model repo.
| call mpas_pool_get_config(configs, 'config_do_restart', config_do_restart) | ||
|
|
||
| if (.not. associated(config_do_restart)) then | ||
| call mpas_log_write('config_do_restart was not found when defining mesh stream.', messageType=MPAS_LOG_ERR) | ||
| ierr = 1 | ||
| else if (config_do_restart) then | ||
| write(stream,'(a)') 'restart' | ||
| else | ||
| write(stream,'(a)') 'input' | ||
| end if | ||
| write(stream,'(a)') 'static' |
There was a problem hiding this comment.
Same 'optional' comment as above.
|
What is the status of this work? @liujake, I saw in the weekly meeting notes that you have another branch that looks to include these changes. |
Hi @jjguerrette, I will update the codes soon according to Jake's suggetions. If you want to use small size files to run your experiments, you can use Jake’s branch now. |
|
Thank you for the info @junmeiban. I will wait until we merge these changes in before using the small file sizes. It sounds like that should not take long. |
|
Optional restart and mpasout workflow with one single code is low priority now give that we have other more pressing work to do. For now, just use JJ's model branch for restart cycling and my branch for mpasout cycling (already tested with cycling experiments at both 120-km and 30-km). |
|
@jjguerrette As you will run/save ensemble analysies/forecasts for EDA, large restart file size for those ensembles could be a burden for you when running more cycles. Disk space may not be a problem for you yet. This two streams workflow should be beneficial to you. |
Currently, we use the restart file for MPAS-JEDI cycling. While we move to higher-resolution cycling experiments, we face the challenge of both memory usage and disk storage. This PR is mainly to figure out the disk storage issue.
The idea is: instead of using restart files, we move to use the file (init.nc type background/analysis file) that keeps only the necessary fields. Also split out static fields into a separate file. In MPAS v6, the capability already exists (Soyoung and Bill: skamaroc@096b5d3). We adapt it in V7 to suit our needs.
Technical details:
static stream: Includes the mesh, some of sfc_input variables(landmask, shdmin, albedo12m, etc) and parameters for gravity wave drag over orography.
da_state stream: Fields are specified in the MPAS-Atmosphere Registry. Includes fields defined in Jake’s doc maps-jedi-da.pdf (section3b) + Soil moisture, soil temperature, etc(these fields are not needed in model initialization but will be used in CRTM)
For cold start, both the static stream file and the input stream file should be set to the “init.nc” file produced by the init_atmosphere core;
For cycling run (FC), the input stream file should be the new da_state stream file that was previously written by the model (and modified/updated in the DA cycle. (Both static and da_state stream need to be specified in the streams.atmosphere file)
For cycling run (DA): use static.nc to read in mesh fields(specified in the streams.atmosphere file) and set config_do_restart = false
How to use this capability:
scripts: /glade/p/mmm/parc/jban/test/pandac/work/JB_fileSizeReduction
results: /glade/scratch/jban/pandac/JB_fileSizeReduction
After implementing the capability, the size of the 120-km init.nc type background/analysis file is about 430M; ~50 variables (the size of original restart file is about 2GB). Memory usage also decreased when we separated the static stream: