Hi all

I am experiencing a problem about partitioning in parallel processing. My fluent jobs die right at the partitioning. I have been doing some experiments on this. And I found out that it might be because of warning about unreference faces in fluent(I get this when I read my mesh file). Normally I just do the reordering to eliminate this warning and partitioning the mesh in single node before running parallel job. The problem is my mesh get biger and contains around 11 million cells now, so I have tried to do the partitioning in a parallel job but it just die at the partition.

Anyone know how why this warning occurs so that it can get fixed? Note that the mesh was successfully meshed in GAMBIT.

If my memory serves me correctly (check me on this...) Fluent did not like partitioning the mesh across an inflow or outflow boundary. Is it possible that you have divided an inflow or outflow boundary between two or more partitions? If so, try re-partitioning so as not to do this. hope this helps...

The unix machine automatically do the auto-partitioning for me. If I request 2 processors then it will divide my mesh into 2 partitions(I guess it use principle axis to do this). But my problem is how to get rid of the unreference faces in the mesh. I just don't know how all the unreferenced faced exist. It seems to me that I have done everything properly in GAMBIT but I still get the warning everytime I read the mesh in fluent. Maybe I missed somethings?

If you have the documentation for fluent, section 31.5.1 of the Fluent 6.3 user's guide covers grid partitioning techniques, both automatic and manual. Also, from reading the documentation, if you're using adaption on a mesh with non-conformal interfaces, you have to partition it manually and save in the serial solver first before loading it for parallel computation.

Hope this helps,

Andy

yeah thanks, the problem was fixed. I know now why I got that "unreferenced faces". They were unreferenced because they related to unspecified mesh in a volume(solid in my case).

[emreaskin]

Solution:

Change the "Green Gauss Cell Based" with the "Least Square Cell Based" method.