And I'm also comparing the above with WRF-EMS 3.4.1.
I have compiled the various WRF versions using last Intel compilers, same libraries and MPI, same configure.wps, configure.wrf options (in particular Linux x86_64 i486 i586 i686, [Xeon (SNB with AVX mods)] ifort compiler with icc (dmpar)).
I'm running a test domain, ARW core, adaptive time step is enabled with these entries in namelist.input:
Code: Select all
time_step = 30, time_step_fract_num = 0, time_step_fract_den = 10, time_step_dfi = 60, use_adaptive_time_step = T, adaptation_domain = 1, step_to_output_time = T, target_cfl = 1.2, target_hcfl = 0.84, max_step_increase_pct = 5, starting_time_step = 20,
Domain resolution is 4 km.
A strange thing happens: WRFV3.6.1 behaves differently from the others with regard to the setting of the time step.
Looking into rsl.out.0000 file I see that the version 3.6.1 always sets a very small time step (average is 15 s).
Instead older versions often use "smarter" (=greater) values (often I see 60 s).
The consequence is that the execution of wrf.exe version 3.6.1 requires 4 times the time it takes for the older versions
I do not know why this adaptive time step has suffered "loss of intelligence" in version 3.6.1.
Does anyone can help?