Scheduled Downtime
On Friday 21 April 2023 @ 5pm MT, this website will be down for maintenance and expected to return online the morning of 24 April 2023 at the latest

Very slow run of real.exe

This post was from a previous version of the WRF&MPAS-A Support Forum. New replies have been disabled and if you have follow up questions related to this post, then please start a new thread from the forum home page.

cyh082

New member
Hello,

I am trying to compile and run WRF model (version 4.2.2) on Linux cluster (CentOS 7).

I used intel 18 compiler with mvapich2.2 library and NetCDF 4.6.1 library (also tried version 4.1.3).

I selected configure option 15.

Actually, I did not have any error when I compiled WRF model.

However, when I tried to run WRF model with multiple nodes, WRF model (real.exe or wrf.exe) ran very slowly (slower than using single node).

Could you give me some advice on the above issue?

Thank you!
 
Hi,
This topic has been moved from the compiling section to the real.exe section, as this is not a compiling issue. Someone will respond to your inquiry soon.
 
Would you please send us your namelist.input to take a look? Also, how many nodes (processors) did you use to run this case?
Please confirm that option 15 is intel compiler and the model was compiled in dmpar mode. Thanks.
 
Hello Ming Chen & kwerner,

I attached "namelist.input" file.

I used 2 nodes, and each node has 28 processors.

Thanks!

Yonghan
 

Attachments

  • namelist.input
    4.4 KB · Views: 68
Yonghan,
All the options in your namelist.input look correct and there is nothing wrong that would slow the model run. The intel compiler and dmpar mode are also fine. I talked to our software engineer and we are suspicious that the slow run of WRF is due to the mvapich library you used. Under certain circumstances, mvapich could run very slow. Would you please try other MPI libraries?
 
Hi Ming Chen,

Yes, as you suggested, I changed MPI library from mvapich to impi, and the problem was resolved.

Thank you for your kind help!

Yonghan
 
Top