[erichu]

Nice to hear that finally works.

However, just another piece of information. The other decompose methods can be left in the decomposeParDict. But you have to choose one and have it properly configured, the other coeffs will not be used and are ignored. At least this is working for me.

Another thing which might be worth spending time is number of CPUs and decomposition direction.
2 CPUs might be faster than 4 in some cases due to slow communication between the cores. Therefore try with 1,2,3,4 CPUs and look at the time.

In some cases the direction also plays a part. I.e 2 CPUs decomposed in (2 1 1) can be faster than (1 1 2) depending on your geometry.

[musahossein]

I started without invoking the mpi. The tank has 73,441 cells. Time step is 0.02, and total time is 12 (so I will need 12/0.02= 600 time steps to complete). When I ran this on my laptop w/o parallel processing, it took 16 hours to complete. When I used parallel processing to use the 4 CPU's that my prosessor has, the computation time was cut down to about 4 hours. So computation time seemed to have reduced linearly. When I checked the performance of the CPUs, it showed them running at 99%-100%. I dont know how much time was spent in fetch, but the fact that the computation time was cut down by 1/4 was a big leap.
I will try your suggestion about the decomposition. However, if my tank is 2 dimensional and the mesh is square -- 271 cells in each direction, will changing the decomposition make a difference? I will try and see.

[nanes]

Originally Posted by musahossein

I have 2 processors. Each processor has 2 CPUs. So I should set subdomain to the number of processors (2) or the number of CPUs (4). Also, I have to run blockMesh and setFields before running mpi, correct? Thanks.

You do not have 2 processors, yuo have one processor with two cores, each core can execute two threads.

http://ark.intel.com/products/47663/...cache-2_13-ghz

The information of gnome-system-monitor is not very accurate.

However MPI processes in total are always 4.

The linear computing time is due to the small number of cells. When the number of cells becomes large (millions) the hyperthreading of your processor does not bring any benefits.

[musahossein]

My processor is Intel® Core™ i3-330M Processor
(3M Cache, 2.13 GHz X4), which makes me think that there are 4 CPUs. You may be right that the genome is interpreting each thread as 1 CPU hence telling me that there are 4 CPUs. I already realize that the laptop is inadequate for the task at hand, because with only 73,000 cells it is taking 4 hours. So I will have to invest in a server which i am in the process of doing.

[musahossein]

Dear all:

I am running OpenFoam/interDyMFoam/sloshingTank2D on a 12 core (dual hexacore) machine. However, the parallell processing gives me the following error:

--> FOAM FATAL ERROR:
Wrong number of processor divisions in geomDecomp:
Number of domains : 12
Wanted decomposition : (4 2 2)

In the decomposeParDict, I have the doman as simple and decomposition as 4 2 2 as shown above. So why am I getting this error?

Any help will be greatly appreciated, Thanks!

[akidess]

Because 4*2*2=16

[musahossein]

you are right. I looked into the source code for gomeDecomp and noted that number of processors must be equal to nx X ny X nz. The manual does not explain that.

Thanks.