Issue wth fluent 6.3.26
 

Hi,
I had already asked the question a week ago or so. I guess, I didn't pose the question properly and hence got no responses...

Here is my situation..

I have a case file and a data file which I am reading into

A) Fluent 6.3.26
B) Ansys 12

In both cases i am using 4 processes. It is a LES simulation.

I have made the mesh in Gambit..

Iniitially, when I ran the fluent, it partitioned using metis and the speed on the both A and B were comparable. After running the simulation for a few residence time, I adapted the grid based on velocity gradient and and after that

case A take about 3 mins per iteration and case B takes about a min

I am not able to figure out why..

here is what I have obsvered, in case A, it says the communication is ethernet and in case B, it says shmem.

and after I adapted the grid, it no longer partions with metis and it uses principal axes ( it gives a warning that there it can't use metis and hence is using principal axes)..

Ansys 12 is installed on a local machine and hence i couldn't use more than 4 processes( quadcore). However 6.3.26 can be run using 16 licenses and when i run it that way, it takes lot longer than what the Ansys 12 takes for 1 time step using just 4 cores.

Any idea as what might be going on..

Regards,
Vishwa

I had similar experience. The problem is related dynamic adaption. When dynamic adaption is specified (e.g. tell fluent to adapt the grid after every 100 iterations), fluent will repartition the grid after every adaption, but the repartition will become very strange due to unknown reason. For example, if I use 20 processes, after the repartition following the adaption, some the partitions do not have any cell in it while some of the partitions have too many cells.

But if do it in the following way, then everything is fine. Do NOT use dynamic adaption, use manual adaption instead. After a number of iterations, adapt the grid manually, repartition the grid manually, make sure the repartition looks good, and then continue the iteration.

If anyone knows how to make it right even with dynamic adaption, please let me know.