Segmentation Fault (WRF V4.5.2)

[joyeede]

On 80 cores (160 logical cores) for 3 km resolution in 1 domain, only for 1 second timestep, there is no segmentation fault and output is successfully been produced. But apart from that, for higher time-step values, there is segmentation fault every time I try to run wrf.exe.
Also, I cannot afford to run the wrf.exe in 1 second time-step because it is taking too long - for example: 2 hours simulation is taking 24 hours to complete.
I have attached the namelist files along with the rsl error file below (for 3 second timesteps).
I would be grateful if someone would help me with this issue. I really want to run the simulation without any faults but in far shorter time.

Thanks in advance.

 

[Ming Chen]

I have a fee concerns about your namelist.input:
(1) for dx =3km, time_step should be 15s or 18s. There is no reason that you can only set time_step = 1s
(2) Is there any special reason that you set etac = 0.01? Its default value is 0.2
(3) This is a big case with grid numbers of 1198 x 1216. This case requires large memory to run.

I would suggest that you run a smaller case first, just to make sure the input data and the options are correctly specified. Then you can move on to run this large case.

 

[joyeede]

Thank you for the suggestions.

Yes, I tried with smaller domain and it works with exactly the same values. Even with higher values of time-steps, it works.
But as for the etac value, it only works when the value is kept at 0.01. For 0.1 and 0.2, fault occurs.

1. Point 1 is solved only after managing Point 3, i.e. after reducing domain size.
2. But as for point 2, does the value of 0.01 for etac affect the run in some way even if the run successfully completes?
3. Also, is there some way to run the large original domain by making any changes in the namelists without causing the fault to occur?

Thanks in advance.

 

[Ming Chen]

The value of etac determines when vertical levels are isobaric, and 0.2 is a well tested and reasonable value. I won't suggest you change it.

For this large case you are running, the issue is more like a memory issue. Please run with more processors, which will give you larger memory. If the issue persists, please talk to your computer manage and manage to obtain more memory.

 

[joyeede]

RAM utilization is just within 15% of the total. So I don't think the issue is with memory.

Just to update you: I have solved the issue. It has to do with epssm. I added epssm=0.2 in the namelist.input. I think it was you have answered in another similar query in the forum with this solution.

Thanks for the help.