search

WRF-CMake / wrf

🌀 The Weather Research and Forecasting (WRF) model with CMake support
Other
44 stars 3 forks source link

mpirun on dmpar binary #8

Closed fronzee closed 5 years ago

fronzee commented 5 years ago

seems that processes does not communicate with each other and the calculation times are the same weather for 1, 6, 8 or more processes.

wrf.exe job finishes ok, but with exactly same timerun independent on core numbers using the wrf-cmake-4.0-dmpar-basic-release-linux

letmaik commented 5 years ago

Speedup depends on many factors. Have you used regular WRF before with the same inputs and achieved better performance? Also, what kind of machine are you running this on and how big is the domain in terms of grid cells?

fronzee commented 5 years ago

It is small initial test domain to know if all the tools are working correctly. 3km grid is 35x35. I havent run it before. I am running it on google cloud compute engine

letmaik commented 5 years ago

Small domains like that probably don't benefit from MPI since the overhead of splitting/combining the computation is too big. I would test it with a larger grid. Also, you haven't mentioned whether you test on multiple MPI-linked nodes, or whether you run on a single VM which has multiple cores (how many?) available.

fronzee commented 5 years ago

Thanks for helping. I will test with bigger domain. I am testing with single VM with 16 cores, however there were not even a slice difference calculating with 1, 6,8 or 16 cores. Is it possible to see if WRF split jobs between cores and how? What should I be looking for?

Thank you

Il giorno Dom 13 Gen 2019, 23:00 Maik Riechert notifications@github.com ha scritto:

Small domains like that probably don't benefit from MPI since the overhead of splitting/combining the computation is too big. I would test it with a larger grid. Also, you haven't mentioned whether you test on multiple MPI-linked nodes, or whether you run on a single VM which has multiple cores (how many?) available.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/WRF-CMake/WRF/issues/8#issuecomment-453869553, or mute the thread https://github.com/notifications/unsubscribe-auth/Ar1xPnyd7K556628RHnitbZO4J2Ag7nCks5vC6x6gaJpZM4Z9VMo .

letmaik commented 5 years ago

Please direct any questions not related to the CMake variant of WRF to the WRF community directly, e.g. http://forum.wrfforum.com/.

dmey commented 5 years ago

@fronzee have you been able to solve your problem? If so I would close this as it does not appear to be a problem related to WRF-CMake.

fronzee commented 5 years ago

No, i couldn't solve it. Maybe it was a problem with mpirun. I compiled from original source (not CMake) with mpich and now it is running.

dmey commented 5 years ago

MPICH is just one implementation of the MPI standard -- mpirun or mpiexec are programs to control the execution of MPI programs. The binary distributions for Linux have also been created using MPICH library (specifically, version 3.2.1). @fronzee could you share your wrf.exe input files as well as the namelist.input so that I can try to run it locally (simply compress and drag and drop the archive in this issue)? Also it would helpful if you could tell me your system configuration (e.g. with linux distro and version, gcc and mpiexec version) and how you run it (e.g. mpiexec ...)-- the more details, the easier and faster will be to sort out. Thanks!

dmey commented 5 years ago

Closing as no response from the user. @fronzee feel free to reopen if the issue persists.