2 Time statistics files

The variable TIMER_Debug, defined in the namcouple (second number on the line below $NLOGPRT keyword), is used to obtain time statistics over all the processors for each routine.

Different output are written (in files named *.timers_xxxx) depending on TIMER_Debug value :

TIMER_Debug=0 : nothing is calculated, nothing is written.
TIMER_Debug=1 : the times are calculated and written in a single file by the process 0 as well as the min and the max times over all the processes.
TIMER_Debug=2 : the times are calculated and each process writes its own file ; process 0 also writes the min and the max times over all the processes in its file.
TIMER_Debug=3 : the times are calculated and each process writes its own file ; process 0 also writes in its file the min and the max times over all processes and also writes in its file all the results for each process.

Note that TIMER_Debug can also be set to -1 to activate the lucia tool that can be used to perform an analysis of the coupled components load balance. More information can be found in the README file in oasis3-mct/util/lucia directory and report mentioned therein.

The time given for each timer is calculated by the difference between calls to oasis_timer_start() and oasis_timer_stop() and is the accumulated time over the entire run. Here is an overview of the meaning of the different timers as implemented by default. ¹³

'total' : total time of the simulation, implemented in mod_oasis_method, i.e. between the end of oasis_init_comp and the mpi_finalize in routine oasis_terminate.
'init_thru_enddef' : time between the end of oasis_init_comp and the end of oasis_enddef, implemented in mod_oasis_method.
'part_definition' : time spent in routine oasis_def_partition.
'oasis_enddef' : time spent in routine oasis_enddef; this routine performs basically all the important steps to initialize the coupling exchanges, e.g. the internal management of the partition and variable definition, the definition of the patterns of communication between the source and target processes, the reading of the remapping weight-and-address file and the initialisation of the sparse matrix vector multiplication.
'grcv_00x' : time spent in the reception of field x in mct_recv (including communication and possible waiting time linked to unbalance of components).
'wout_00x' : time spent in the I/O for field x in routine oasis_advance_run.
'gcpy_00x' : time spent in routine oasis_advance_run in copying the field x just received in a local array.
'pcpy_00x' : time spent in routine oasis_advance_run in copying the local field x in the array to send (i.e. with local transformation besides division for averaging).
'pavg_00x' : time spent in routine oasis_advance_run to calculate the average of field x (if done).
'pmap_00x'/'gmap_00x' : time spent in routine oasis_advance_run for the matrix vector multiplication for field x on the source/target processes.
'psnd_00x' : time spent in routine oasis_advance_run for sending field x (i.e. including call to mct_waitsend and mct_isend).
'wtrn_00x' : time spent in routine oasis_advance_run to write fields associated with non-instant loctrans operations to restart files (see section 5.2 for details).
'wrst_00x' : time spent in routine oasis_advance_run to write fields to restart files (see section 5.2 for details).

Footnotes

... default.¹³: Many other measures can be obtained by defining the logical local_timers_on as .true. in different routines or by implementing other timers. Of course, OASIS3_MCT and the model code then have to be recompiled.