Hi,

I have a 2d domain that i am running parallel on a cluster. I am interested in monitoring field variables in many points. For this i am using the monitor/surface/area-wighted-average for my different points. I want to gather the data for every time step.

Now, the problem is that fluent REALLY saves the data in the file every time step. I notice that my calculation is now 50-70 % as fast as without monitoring. I can see, using top, that this is due to one node having a cortex and other process stealing CPU from the solver.

How can i make fluent save the data in memory buffer before writing it to the files (lets say write the file every 200 time steps). ???

as it is now is just a waste..

Thanks!

I am running fluent with the -g flag. So i am not using graphics at all.

I do not why there is a cortex process at all. I made sure it is not from some old crashed run of fluent.

Did anyone experience this before?

Suggestions for a solution?

Hi again,

I now know that it is not due to the writing of each time step that i get a reduction in performance. If i do a test of writing, say every 60 time step, i still get the reduction in speed.

i can see from "top" that a cortex process is fluctuating between 0 and 70 % of CPU on one node during 1 time step.

Thus it seems that only the presence of my monitor points is enough to cause this problem.

any ideas?