Segmentation Fault when using YSU PBL scheme and ETA/Ferrier for mp scheme

[luizfcs]

I succefully run WRF using WRF Double-moment 6-class (WDM6) as a microphisic scheme and Mellor-Yamada-Janjic (MYJ) as PBL scheme.
But when i try to use ETA Ferrier microphisic scheme it crashes, due to segmentation fault.
Also when i try to uso YSU as PBL scheme, even with WDM6, it crashes due to segmentation fault. I hve no idea what coul possibly be causing this, any help?

Runing on EC2, 32 processors, mpirun np4 -bind to core:8

 

[Ming Chen]

What is the error message? Did the model crash immediately?

 

[luizfcs]

It crashes imediately, the error message: is segmentation fault (11)

After some more tests i narrow it down the problem:

I was able to run with a smaller domain 100x100 with the same mpi passing ( 4 process, 8 cores per process)

I tryed again with a larger domain 160x160, but i was only able to succefully run the simullation with 16 process and 1 core per process.

On my last try, i expanded my domain to 240x240, but with this gridsize i was unabble even to run previous parametrizations MYJ and Goddar, wich was running with a 200x200 grid)

My question now is if its only a siple matter to use more processors, or if my parallell strategy isn't right.

I am running the WRF on an AWS EC2 wrf-arm c7g.8xlarge with 32 VCPU, 64 GB ram memory. And i am usin dmshare strategy (don't know if i should use SM ou DM + SM with VCPU from AWS)

 

[Ming Chen]

Based on the information you posted, I believe this is a memory issue. I wonder whether you can compile WRF in dmpar mode, then run the case with more number of processors, which will give you larger memory.
I am not quite familiar with AWS. If this issue persists, please contact AWS manager to get larger memory.

 

[luizfcs]

Thank you for the support!

It turns out it was really a memory issue. I tryied some more tilying strategyies and mannaged to run thw 240x240 grid with 32 process. To max perform the elapsed time rate of each time step i am using 1 processor for each process. I'm on the limit of the AWS machine now, but at least it works.

 

[Ming Chen]

Thanks for the update. I am glad it works for you!