Hi everyone, I've been looking everywhere on the forum and in Fluent documentation, but still can't find any solution to my problem : I run 2 parallel applications (Linux) on 2 dual core processors( cpu1, cpu2, cpu3 and cpu4 to name the different cores): If the 1st calculation is launched with : fluent 3ddp -t2, it will use cpu1 and cpu2. If I run a second calculation with the same command fluent 3ddp -t2, Fluent assigned the 2nd calculation to also cpu1 and cpu2 instead of using cpu3 and cpu4. However with a parallel job running (on cpu1 and 2), if I run a serial case it will automatically use the cpu3.

So How can I specify to fluent to use 1 pair of cores per parallel job instead using the first two one. I guess it has to do with the hostfile thing...but not very clear I think.

Thanks for any help. Sam

write a hosts file (example in manual) then use switch fluent 3d -cnf=hostfile.txt and these nodes will be called to spawn

Hi Sam, on dual core you have to set the affinity of each compute node manually. After starting Fluent and reading the case file use the scheme command (set-affinity "n0 p0 n1 p1 ...") where n0, n1, ..., are your compute nodes and p0, p1 ...are your core-ids.

Max

This is due to an issue with FLUENT when you try to run two FLUENT jobs on the same machine. The solution from max is good, but you need to set it after the case file is read, because during case file read affinity is always been reset. This problem is supposed to be fixed in FLUENT6.4.

You may have another fix by putting the following lines in a file called ~/.fluent ;;;;;;;;;;;;;;;;;; (set! set-affinity

(lambda (str) (display "User defined set-affinity\n")) ) ;;;;;;;;;;;;;;;;;;

Good luck!

Hi ,

Thanks again for the answers, I'm going to try that and give you the feedback !

Cheers