Tile Strategy is not specified.

[svangorder]

Hi,

I get the following in my wrf.exe standard output file.

Tile Strategy is not specified. Assuming 1D-Y
WRF TILE 1 IS 160 IE 318 JS 1 JE 178
WRF NUMBER OF TILES = 1

I'm running WRF 4.0.3 distributed with the COAWST modeling system.
This is a 12-processor MPI run and I have

nproc_x = 4
nproc_y = 3

in the &domains namelist.
Full namelist.input attached.

Can someone please tell me what this means?

Thanks,
Steve


View attachment namelist.input

 

[kwerner]

Hi Steve,
So for your domain set-up, you have
e_we = 637
e_sn = 534
WRF domains are 1-based, which means that your domain is actually 533 x 636. Since you have 12 processors, this divides your domain into a 4x3 grid with tiles of size 158x177 (some rounding is done, and adjustments are made when the number is not perfectly divisible). So this particular print-out looks like it's for the 2nd tile over from the start of i (SW corner), and the first tile in the j-direction. Therefore this line:

WRF TILE 1 IS 160 IE 318 JS 1 JE 178

actually means:
for this particular tile,
i-start = 160
i-end = 318
j-start = 1
j-end = 178

Your decomposition should be:
i-direction
1-159
160-318
319-477
478-636

j-direction
1-178
179-356
357-534

Hope that helps a little!

 

[svangorder]

Hi Kelly,

Yes those are the grid is "tiles" for splitting up the computation amongst the 12 processors.

There are 12 entries like that in the standard output - one for each processor.
They ALL say

Tile Strategy is not specified. Assuming 1D-Y
WRF NUMBER OF TILES = 1

The only difference being the line that gives the values of IS, IE, JS, JE

So unless I'm miss-interpreting, here a "WRF TILE" seems to refer to some kind of sub-tile of each computational tile?
Do you know what that is? ... or maybe I am just confused.

Steve

 

[davegill]

Steve,
You are correct. If this were an OpenMP build, the "WRF Tile" information would refer to a sub-tiling of the MPI rank's patch.

It would be better if the following line in the model output did not appear unless we were running OpenMP:

Tile Strategy is not specified. Assuming 1D-Y

Again, this printout should be removed from the output, but the line is currently there. Without OpenMP, it signifies nothing more than a reminder that the computational area on each MPI rank is the work that is handled by a single MPI task.

The next line(s) are useful for MPI and for OpenMP. If this is an OpenMP build, there will be as many lines with "WRF Tile" as there are OpenMP threads. When this is an MPI-only build, then the single line with the tile size (the information on the "WRF Tile" line) is identical to the patch size (the information a few lines up that, that is on the "ips,ipe,jps,jpe" line).

 

[svangorder]

Thanks for your help Kelly, that clears it up!

Best,
Steve