Changeset 3394 for trunk

Timestamp:

Jul 8, 2024, 11:48:37 PM (6 months ago)

Author:

jbclement

Message:

PEM:
Updates and improvements related to the last modifications of the launching script.
JBC

Location:

trunk/LMDZ.COMMON/libf/evolution

Files:

• changelog.txt (modified) (1 diff)
• deftank/README (modified) (1 diff)
• deftank/clean.sh (modified) (1 diff)
• deftank/lib_launchPEM.sh (modified) (1 diff)
• pem.F90 (modified) (4 diffs)

trunk/LMDZ.COMMON/libf/evolution/changelog.txt

@@ -395 +395 @@
 == 02/07/2024 == JBC
 Addition of the possiblity to launch the 1D model in the launching script + correction of small errors.
+
+==08/07/2024 == JBC
+Updates and improvements related to the last modifications of the launching script.

trunk/LMDZ.COMMON/libf/evolution/deftank/README

@@ -6 +6 @@
       (i)   n_mars_years, n_earth_years -> the number of Mars/Earth years to be simulated in total;
       (ii)  nPCM_ini -> the number of initial PCM runs (at least 2);
-      (iii) nPCM -> the number of PCM runs between each PEM run (usually 2).
+      (iii) nPCM -> the number of PCM runs between each PEM run (usually 2);
+      (iv)  dim -> the dimension of the model (1 for 1D, any other values stand for 3D).
   The script can take an argument:
       - If there is no argument, then the script initiates a PEM simulation from scratch.
       - If the argument is 're', then the script relaunches an existing PEM simulation. It will ask for parameters to know the starting point that you want to. 
+  The script works only with the job scheduler SLURM to submit chained jobs.
 
 # liblaunchPEM.sh:
   Library of bash functions used by the PEM launching script "launchPEM.sh".
 
-# jobPCM.slurm:
-  Bash script file to run the PCM with a SLURM job. The name of the PCM executable file should be adapted. The header corresponds to the ADASTRA supercomputer and should be changed for other machines.
+# PCMrun.job:
+  Bash script file to run the PCM with a SLURM job. The name of the PCM executable file should be adapted. The header corresponds to the ADASTRA supercomputer and should be changed for other machines. In case of 1D, the SBATCH headers are omitted.
   The path to source the arch file should be adapted to the machine.
+  The execution line should also be adapted according to the set-up.
 
-# jobPEM.slurm:
-  Bash script file to run the PEM with a SLURM job. The name of the PEM executable file and Reshaping executable file should be adapted. The header corresponds to the ADASTRA supercomputer and should be changed for other machines.
+# PEMrun.job:
+  Bash script file to run the PEM with a SLURM job. The name of the PEM executable file and Reshaping executable file should be adapted. The header corresponds to the ADASTRA supercomputer and should be changed for other machines. In case of 1D, the SBATCH headers are omitted.
   The path to source the arch file should be adapted to the machine.
+  The execution line should also be adapted according to the set-up.
+  The PEM executable can have an optional argument to specify the SLURM job ID in order to detect the job time limit and deal with it.
 
 # run_PEM.def

trunk/LMDZ.COMMON/libf/evolution/deftank/clean.sh

@@ -10 +10 @@
 rm -f data_PCM_Y*
 rm -f info_PEM.txt
-find . -type f -name "jobPCM*.slurm" ! -name "jobPCM.slurm" -delete
+find . -type f -name "PCMrun*.job" ! -name "PCMrun.job" -delete
 rm -f *.out
 rm -f kill_launchPEM.sh

trunk/LMDZ.COMMON/libf/evolution/deftank/lib_launchPEM.sh

@@ -39 +39 @@
     LC_NUMERIC=en_US.UTF-8
 
-    if [ $dim -ne 1 ]; then # 3D model works only with SLURM for now (because of time limit)
-        islurm
+    if [ $dim -ne 1 ]; then # 3D model works only with SLURM for now (because of chained job submission in the script + time limit management)
+        is_slurm
     fi
     if [ -v n_mars_years ] && [ ! -z "$n_mars_years" ]; then

trunk/LMDZ.COMMON/libf/evolution/pem.F90

@@ -234 +234 @@
 real            :: timestep        ! Timestep [s]
 character(20)   :: job_id          ! Job id provided as argument passed on the command line when the program was invoked
+logical         :: timewall        ! Flag to use the time limit stopping criterion in case of a PEM job
 integer(kind=8) :: cr              ! Number of clock ticks per second (count rate)
 integer(kind=8) :: c1, c2          ! Counts of processor clock
@@ -285 +286 @@
 call system_clock(count_rate = cr)
 call system_clock(c1)
+timewall = .true.
 timelimit = 86400 ! 86400 seconds = 24 h by default
 if (command_argument_count() > 0) then
@@ -317 +319 @@
         timelimit = minutes*60 + seconds
     endif
+else
+    timewall = .false.
 endif
 
@@ -1006 +1010 @@
     if (i_myear >= n_myear) stopPEM = 6
     call system_clock(c2)
-    if (real((c2 - c1)/cr) >= timelimit - antetime) stopPEM = 7
+    if (timewall .and. real((c2 - c1)/cr) >= timelimit - antetime) stopPEM = 7
     if (stopPEM > 0) then
         select case (stopPEM)