Number of Meshes exceeding number of MPI Processes #14756
-
|
Hello, below is a snippet of my fds file &MESH ID='Mesh02', IJK=48,48,60, XB=235.0,247.0,129.0,141.0,0.0,15.0, MPI_PROCESS=0 / and below is the execution I am trying to run fds_local -p 8 -o 2 DF1.fds My PC does have the required cores and processors for this and I am getting Error 115. ERROR(115): Number of meshes exceeds number of MPI processes. Set MPI_PROCESS on each MESH line so that each MESH is assigned to a specific MPI process (CHID: DF1) Is there something I am missing? I have 8 MPI Processes (0-7) and they are numbered in order. Do the meshes that are assigned the same process have to be physically adjacent in the model? The intent of this was to balance out the cell sizes in the different meshes. Any help is greatly appreciated. Thanks, |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
I added: to your MESH lines, and the case began to timestep with the current source. Are you running the current version of FDS? If you have 16 physical cores, you would see faster speed up using 16 meshes rather than 8 meshes and 2 OpenMP processes. -o 2 might speed things up 40 - 50 % vs just -p 8, but -p 16 should be close to 100 % faster than just -p 8. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks I had a look at it appears the FDS file had duplicated within itself, snippet below &OBST ID='Obstruction', XB=340.5,341.5,149.5,151.5,2.0,3.0, RGB=153,204,255/ &HEAD CHID='DF1'/ &MESH ID='Mesh02', IJK=48,48,60, XB=235.0,247.0,129.0,141.0,0.0,15.0/ So there are more meshes in the file. Not sure what's caused the duplication of the fds file when it was written but ill try cleaning up the fds file. |
Beta Was this translation helpful? Give feedback.
I added:
&HEAD CHID='test'/
&TIME T_END=0.0001/
to your MESH lines, and the case began to timestep with the current source.
Are you running the current version of FDS?
Do you have any MESH lines someplace else in your input file?
If you have 16 physical cores, you would see faster speed up using 16 meshes rather than 8 meshes and 2 OpenMP processes. -o 2 might speed things up 40 - 50 % vs just -p 8, but -p 16 should be close to 100 % faster than just -p 8.