Large parellel run with small-n outer nest

[jamesrup]

Hi all,

I'm having an issue where I need to use many CPU cores for a very high-resolution inner nest but I'm telescoping down with 3 nests, and the outermost domain is quite small in nx,ny. Due to that, I'm getting an error that cell width is too small for this nest, with the message "Minimum decomposed computational patch size, either x-dir or y-dir, is 10 grid cells."

It would seem that my only options are to

1. Increase nx,ny of the outer domain or
2. Reduce the number of nodes/CPUs I'm using

Those are both very undesirable. Any workarounds?

Thanks!
James

 

[kwerner]

James,

Yes, unfortunately you are going to need to increase the size of d01. However, this shouldn't cause too much issue because most of the computation expense is taking place in the inner domain that is much larger and higher resolution. It really shouldn't increase the computation expense much at all to simply make that domain larger.

 

[jamesrup]

Okay, thanks for the reply Kwerner.

I'm really hesitant to invest the large additional number of nx,ny points, but thinking about this more, I guess the much longer time step for that domain this keeps the added overall cost quite minimal...

James