Query on HYSPLIT MPI version

[vikrant_sapkota]

Hi,
Currently I am using HYSPLIT linux version for simulating BC concentration in Nepal using Egar HTAP emission inventory. Since the inventory is gridded and consists of more than 70000 grid points, the runtime for 8 day HYSPLIT has been over 96 hours when running in serial mode.

Parallel version using MPICH does not show much improvement in the run completion time. It seems to us that the parallel runs are correct; it shows all cores being run when checking via top command and creates all the MESSAGE files corresponding to the number of processors used. However, the runtime is the same as serial version if not slower. Please find attached a control file, emitimes file, one message file of a sample run with 300 gridded points. The run time for this trial was ~2 hours for both serial and parallel runs. My system details is as follows " Linux opensuse 4.4.140-62-default #1 SMP Tue Jul 17 12:14:34 UTC 2018 (951943d) x86_64 x86_64 x86_64 GNU/Linux". I have use hycm_std executable for the mpi run.

Any help from your side regarding the matter would be greatly appreciated.




Regards,
Vikrant Sapkota

CONTROL.txt

(9.63 KiB) Downloaded 243 times

The attachment EMITIMES.txt is no longer available

The attachment CONTROL.txt is no longer available

[Fabio.Dioguardi]

Hi,

I am facing a similar problem.
I have compiled libraries and the source code, the latter by modifying the Makefile to use mpich and produce hycm_std (see attached). When I run it with mpirun, it looks like it is running in parallel using the prescribed number of processes (40 in my case, and I can see in the run folder all the 40 WARNING and MESSAGE files), but the time it is taking is comparable with exactly the same simulation run in serial mode.
Any idea on why is this happening?

Thanks
Regards
Fabio