[pkr]

Hi,

I am looking at nonNewtonianIcoFoam under incompresible. It invokes PCG solver to solve Ax=b kind of equation. I am interested in increasing the size of the matrix such that each iteration of matrix-vector multiplication takes large amount of time to complete.

Currently, I managed to increase the n_cells to ~14K by running the fineMesh utility. But the matrix-vector multiplication just takes 0.1 ms per invocation. Please suggest how can I increase upon the computation inside matrix-vector multiplication.

Thanks!!!

[Dennis]

Can you elaborate on why you want PCG to take longer? At the moment you still have a relatively small system to solve, so increasing the amount of dofs can increase the time somewhat. Creating a badly conditioned system will also increase your PCG solve time, but this is probably difficult as OpenFOAM will try to avoid that for obvious reasons.

[pkr]

Originally Posted by Dennis

Can you elaborate on why you want PCG to take longer? At the moment you still have a relatively small system to solve, so increasing the amount of dofs can increase the time somewhat. Creating a badly conditioned system will also increase your PCG solve time, but this is probably difficult as OpenFOAM will try to avoid that for obvious reasons.

Thanks for your response. I am evaluating PCG on many-core/multi-core systems. The benefit can only be seen if the computations are at higher side. But in the current scenario, the work done per iteration is low and the number of iterations are high. So I am looking for increasing the compute time for each iteration.

Can you please suggest the way of increasing the amount of dofs?

[CST]

Hey pkr,

how are you doing with your evaluation of PCG-Solvers on multi-core-systems?
Do you compute on just 1 machine or in a network on more than one?

By looking around in the forum I saw this Thread (named 'Cluster OpenFOAM [Solved]'):

http://www.cfd-online.com/Forums/ope...am-solved.html

Have you seen some problems with computing the pressure with PCG-algorithm on more than 1 machine, like it is said in post nr. 6?


With best reagrds,

CST