when openning the conpiled udf library in the parallel version? (it takes more than half an hour, may be much more than that) JUST NO ERROR MESSAGE, NO RESULTS

Also when I set up the problem in the parallel version fluent, it seems slow too! I just tried it between two P4--512M machines which are connected by a 10Mb network. I don't think it will take such a long time to setup this 60,000-mesh problem, even though it's a 3d project. Who could be so kind to give a explanation? TIA

Hi,

Parallel UDF sometimes takes a long time. Because the cell(or face) loop loops twice in case of two CPU . So to avoid this kind of thing ,you have to remove some waste loop or have to do special coding for parallel(ask Fluent's support engineer).

If you think reading your grid is slow, the reason is maybe auto partition function. In case of this, I recommend the following procedure.

-All bundary conditions and so on are defined in serial Fluent. And then partitioning .

-Save case and data before exit serial Fluent.

-Booting parallel Fluent ,read the case and data.

I think that 60,000 mesh is too much small to parallel processing specially the cluster parallel. It is not efficient so much.

Regards,

Keil

I have seen many users have unnecessary tendency to use parallel version of fluent even without checking the efficiency.

Unless during the development of the flow, there is increase of computing (injecting particles/bubble break up), 60000 cells with K-e models (or even RSM) is not a huge domain to run parallel.

Regards, phd_y2k2001@yahoo.com

Thanks for your both suggestions. But I have a question when I open the case file in parallel Fluent: Cause I have a UDF inlet boundary, and you know the directory structures used are different between serial Fluent and parallel Fluent. When you do partition in the serial Fluent for a UDF-inlet problem, the udf files you use is in the directory, for example, \3d. Then you save and recall this saved case file in parallel Fluent. But at this time, what you are opening is the files in \3d_host. Does it matter?

As for the efficiency, how could I know it before I compare them in different cases? Actually, I really appreciate your experienced guys told me a lot, but this is my first time to explore such kind of funciton of the Fluent, I will gain much more by trying parallel Fluent than just saving time.

Thanks a lot,

Hi Knick, I need a small favor from you . I am currently trying to solve a pretty big problem too . I need your help in knowing the steps to do parallel processing . If you could please help me with that it will be great. Thanks a lot, Taher.

Hi, Taher: I'd be glad if I could help. Here is a useful link, you may follow its instruction to get your parallel calculation work: http://www.fluent.com/support/instal...infaq/rshd.htm