code_saturne User's Forum

Filippo, the time it will take depends more on the size of the mesh (quantity of elements) and the "quality" of it, rather than on the physical dimensions, i.e. it will take longer to calculate a 1m beam discretized with 10 elements than the same beam discretized in 4 elements (asuming the mesh converges). About the quality, a mesh will converge faster if the elements are regular, homogeneous in size and shape, with smooth size variations from cell to cell, and preferably if the mesh is structured. I can tell you that good mesh of 10^6 elements has converged faster for me than a bad 10^5 mesh.
Having said that, I've been running a steady flow simulation with a mesh of 10^6 elements, mostly structured (but not high quality), laminar flow, in a low end hardware (Sempron processor with 1.9GHz, only one core, 32bit Mandriva system) and I got quite surprised to see that each step took only 2mins or less to finish and using no more than 1.5GiB RAM. With a modern multicore precessor that time will drop amazingly.
Hint: make a run of a case, with different mesh densities, and check regualrly the "listing" file. At the end of each step (provided you have configured it so) you'll see the time it took to finish that step, and after the 100th step (usually) the shown time stabilizes.

ok thank you, i got it.
well now I have set up a grid of probes to get data about pressure and velocity of the wind flow but seems that this grid cause errors in calculation: with the probes the calculation doesn't run...
It's a lot of probes like 120 but i need them in order to find the pressure coefficients induced by wind.
I get the errors in the listing file...
I apologize for the continual questions...
Thank you again,
Filippo

Hi Filippo,
The number of probes, ncapt, is limited by the parameter ncaptm = 100.
You could try changing this to 120 (~ line 132 in paramx.h then place it in SRC folder), not sure if this will have an effect (100 was probably chosen for a reason), you can try anyway!
 
Hope that helps,
James

Hello,
You can change ncaptm in paramx.h in the source code and you must recompile and reinstall the whole code (copy paramx.h in SRC folder will not have any effect). The limitation ncaptm=100 was chosen because the memory allocation is static and in most of case 100 probes are sufficient.
 
JF
PS : maybe 1D profiles are better suited for your case.