Changes between Initial Version and Version 1 of ticket/1704


Ignore:
Timestamp:
2016-03-23T18:18:09+01:00 (5 years ago)
Author:
lovato
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • ticket/1704

    v1 v1  
     1 
     2This wiki relates to the ticket on NEMO reproducibility fails with land domains exclusion (#1704) 
     3 
     4'''Author''': Tomas Lovato  
     5 
     6'''Institution''': CMCC 
     7 
     8'''Proposed actions''', after the confirmation from other ST members: 
     9- revise the code and the log of mppini subroutines 
     10- evaluate the modified tool for offline mpp decomposition 
     11 
     12'''Additional details not included in the ticket''' 
     13 
     14I tried to use configuration 4 (ORCA2_LIM_PISCES) to test the reproducibility for a run with a total of 88 PEs (80 ocean+8 land) and one with only the ocean domains (80 PEs).  
     15Here below a diff file of changes used in the test. 
     16 
     17 
     18{{{ 
     19@@ -385,7 +386,7 @@ if [ ${config} -eq 4 ] ;  then 
     20     . ./all_functions.sh 
     21     . ./prepare_exe_dir.sh 
     22     JOB_FILE=${EXE_DIR}/run_job.sh 
     23-    NPROC=16 
     24+    NPROC=88 
     25     if [ -f ${JOB_FILE} ] ; then \rm ${JOB_FILE} ; fi 
     26     cd ${EXE_DIR} 
     27     set_namelist namelist_cfg nn_it000 1 
     28@@ -393,9 +394,9 @@ if [ ${config} -eq 4 ] ;  then 
     29     set_namelist namelist_cfg nn_fwb 0 
     30     set_namelist namelist_cfg ln_ctl .false. 
     31     set_namelist namelist_cfg ln_clobber .true. 
     32-    set_namelist namelist_cfg jpni 4 
     33-    set_namelist namelist_cfg jpnj 4 
     34-    set_namelist namelist_cfg jpnij 16 
     35+    set_namelist namelist_cfg jpni 8 
     36+    set_namelist namelist_cfg jpnj 11 
     37+    set_namelist namelist_cfg jpnij 88 
     38     set_namelist namelist_cfg nn_solv 2 
     39     set_namelist namelist_top_cfg ln_trcdta .false. 
     40     set_namelist namelist_top_cfg ln_diatrc .false. 
     41@@ -425,16 +426,16 @@ if [ ${config} -eq 4 ] ;  then 
     42     export TEST_NAME="REPRO_2_8" 
     43     . ./prepare_exe_dir.sh 
     44     JOB_FILE=${EXE_DIR}/run_job.sh 
     45-    NPROC=16 
     46+    NPROC=80 
     47     if [ -f ${JOB_FILE} ] ; then \rm ${JOB_FILE} ; fi 
     48     cd ${EXE_DIR} 
     49     set_namelist namelist_cfg nn_it000 1 
     50     set_namelist namelist_cfg nn_itend 75 
     51     set_namelist namelist_cfg ln_clobber .true. 
     52     set_namelist namelist_cfg nn_fwb 0 
     53-    set_namelist namelist_cfg jpni 2 
     54-    set_namelist namelist_cfg jpnj 8 
     55-    set_namelist namelist_cfg jpnij 16 
     56+    set_namelist namelist_cfg jpni 8 
     57+    set_namelist namelist_cfg jpnj 11 
     58+    set_namelist namelist_cfg jpnij 80 
     59     set_namelist namelist_cfg nn_solv 2 
     60     set_namelist namelist_top_cfg ln_trcdta .false. 
     61     set_namelist namelist_top_cfg ln_diatrc .false. 
     62 
     63 
     64}}} 
     65 
     66The test for reproducibility failed. 
     67 
     68I made the following change in the mppini_2.h subroutine to modify the 
     69exclusion criteria, which now consider a domain as land when all points 
     70within the region (inner+overlap) are zero (I also modified the offline 
     71decomposition tool accordingly): 
     72 
     73{{{ 
     74isurf = 0 
     75- DO jj = 1+jprecj, ilj-jprecj 
     76- DO ji = 1+jpreci, ili-jpreci 
     77+ DO jj = 1, ilj 
     78+ DO ji = 1, ili 
     79IF( imask(ji+iimppt(ii,ij)-1, jj+ijmppt(ii,ij)-1) == 1) 
     80isurf = isurf+1 
     81END DO 
     82END DO 
     83}}} 
     84 
     85 
     86After these modifications, I rerun SETTE using the configuration 4 test 
     87always using 88 PEs, but this time the decomposition was 81 ocean + 7 
     88land. 
     89The reproducibility test passed. 
     90 
     91 
     92 
     93