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

WRF4.5.1 takes 50min for a 6min case in WRF4.2.2

Hui Liu

New member
I use the same ERA5 data to run wrf4.5.1 and wrf4.2.2, with the same namelist.input, same machine and computing resources (nodes=2, ppn=20).
dx = dy = 30000, time_step = 90, e_we = 180,e_ns = 100.
wrf4.5.1 takes 50min to run 1day, while wrf4.2.2 takes only 6min.
What makes the difference?
If wrf4.5.1 takes such long time, it would not be suitable for any optimization research.
 

Attachments

  • namelist.input
    3.6 KB · Views: 2
  • namelist.wps
    765 bytes · Views: 0
  • rsl422.out.0000
    517 KB · Views: 3
  • wf451.0000
    3.4 MB · Views: 2
I looked at the files rsl422.out and wf451. It seems that wf451.0000 is output from an initial run, while rsl422 is from a restart run.
Please let me if I am wrong.

Can you run the same case with the same settings and see whether there is large difference in the computation between them?
 
I have rerun the case.
WRF4.5.1 takes around 60mins. I have run it twice. (10:19-11:38, wrf451_1.0000) (13:33-14:42,wrf451_2.0000)
WRF4.4.2 takes 6mins.
WRF4.3.3 takes 6mins.
WRF4.2.2 takes 6mins.
 

Attachments

  • wrf451_2.0000
    3.3 MB · Views: 0
  • wrf451_1.0000
    3.4 MB · Views: 3
  • rsl442.out.0000
    84.8 KB · Views: 2
  • rsl433.out.0000
    83.9 KB · Views: 0
  • rsl422.out.0000
    83.6 KB · Views: 1
I guess you are running WRFV4.5.1 using only one processors. Please let me know if I am wrong. I ask tis because I found that
"WRF NUMBER OF TILES = 1" in your wrf451_1.0000 log file.

This is related to other issues:
(1) How did you compile WRFv4.4.2 and WRFv4.5.1?
(2) How did you run them respectively? Did you use the same number of processors?
 
I don't know whether my WRFV4.51 is actually running with only one processors, but I submit the task with 2 nodes and 20 processors per nodes(ppn),
so it should be 40 processors in total. I have tried more nodes or processors, even WRF4.4.2 will slow down. So I chose 2 nodes for best speed.
(1) I compile WRFv4.2.2 and WRFv4.5.1 in the same way. Compiler choice: 15 , all other packages and environments are the same.
I attached the compile.log and configure.wrf for them. (name was changed, it seems that *.wrf file cannot be attached)
(2) I run them in the same way as I mentioned above. I 'qsub wrfpbs.pbs' to submit the tasks. I also attached the file.
 

Attachments

  • compile442.log
    828.4 KB · Views: 1
  • compile451.log
    61.6 KB · Views: 2
  • configure442.log
    23.6 KB · Views: 0
  • configure451.log
    23.2 KB · Views: 0
Last edited:
Here is the content of my wrfpbs.pbs file:
#!/bin/bash
#PBS -V
#PBS -q normal
#PBS -N HW
#PBS -j oe
#PBS -l nodes=2: ppn=20
#PBS -l walltime=500:00:00
cd $PBS_O_WORKDIR
mpirun -np 40 -machinefile $PBS_NODEFILE ./wrf.exe
 
Top