CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   results with parallel processing (

Muhammad Shakaib December 29, 2005 00:07

results with parallel processing
I am solving an unsteady problem using linux cluster at six nodes. For comparison I started the solution (first few iterations)using the cluster and then started the solution using single computer. There was some difference in the printed residual values. Is this usual or something wrong has happened. I am using fluent 6.2.16.


Hubert Janocha December 29, 2005 01:56

Re: results with parallel processing
It is possible that the residuals vary a little bit. When using domain decomposition you will introduce (in some way) an other iterative process, because the boundaries on the domain borders are only virtually, and are not fix. They will be updated e.g. each Iteration. So you have a weak disatvantage in convergence rate/behauviour in parallel mode. But this disatvantage is very small, it depends strongly on the solver settings, and implementation for parallel mode.

Because of this I would be suprised when you get the same residuals.

All times are GMT -4. The time now is 10:10.