parallel results different depending on decomposition strategy

franzisko · November 3, 2009, 10:52

hello,
I performed a simulation using simpleFoam and for different parallelization strategies: number of processors, simple/hierarchical/... I get slightly different results. Is there anyone which knows it is ok, or if there is any problem. The difference seems to be small (more or less 1% on drag coefficient for the stationary solution). Where is the part depending on the decomposition method?

thanks a lot,
Francesco

madad2005 · November 4, 2009, 03:07

Hi Francesco,

I've been having some wider-spread issues with using foam in parallel. I'm assuming the differences you are seeing are due to decomposition of the domain each time with different regions having different ghost cells. 1% does appear to be a lot though. Interestingly, I'm finding that when attempting to run foam in parallel, it appears it is trying to run N number of serial jobs and is killing the memory on my machine. Have you seen similar behaviour?

Regards,

Adriano

madad2005 · November 4, 2009, 05:35

Well, I finally resolved my problem. I was missing the -parallel command out when executing the parallel job.

franzisko · November 4, 2009, 06:37

hi adriano,

thanks for your suggestion concerning ghost nodes. I think 1% is a bit too much, too. however, the fields are very similar each other, hence I hope there is no serious problem

intersting behaviour missing -parallel...

bye, francesco

November 3, 2009, 10:52	parallel results different depending on decomposition strategy	#1
franzisko New Member Francesco Join Date: Jul 2009 Posts: 13 Rep Power: 16	hello, I performed a simulation using simpleFoam and for different parallelization strategies: number of processors, simple/hierarchical/... I get slightly different results. Is there anyone which knows it is ok, or if there is any problem. The difference seems to be small (more or less 1% on drag coefficient for the stationary solution). Where is the part depending on the decomposition method? thanks a lot, Francesco

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Different Results from Fluent 5.5 and Fluent 6.0	Rajeev Kumar Singh	FLUENT	6	December 19, 2010 11:33
MPICH Parallel Run Error on the Windows Server2003	Saturn	CFX	3	August 29, 2006 08:42
results with parallel processing	Muhammad Shakaib	Main CFD Forum	1	December 29, 2005 00:56
CFX-5.7 MPICH Parallel Problem (Output of Results)	James Date	CFX	7	February 15, 2005 16:03
parallel mode - small problem?	co2	FLUENT	2	June 1, 2004 23:47

November 4, 2009, 03:07		#2
madad2005 Senior Member Anonymous Join Date: Mar 2009 Posts: 110 Rep Power: 17	Hi Francesco, I've been having some wider-spread issues with using foam in parallel. I'm assuming the differences you are seeing are due to decomposition of the domain each time with different regions having different ghost cells. 1% does appear to be a lot though. Interestingly, I'm finding that when attempting to run foam in parallel, it appears it is trying to run N number of serial jobs and is killing the memory on my machine. Have you seen similar behaviour? Regards, Adriano

November 4, 2009, 05:35		#3
madad2005 Senior Member Anonymous Join Date: Mar 2009 Posts: 110 Rep Power: 17	Well, I finally resolved my problem. I was missing the -parallel command out when executing the parallel job.

November 4, 2009, 06:37		#4
franzisko New Member Francesco Join Date: Jul 2009 Posts: 13 Rep Power: 16	hi adriano, thanks for your suggestion concerning ghost nodes. I think 1% is a bit too much, too. however, the fields are very similar each other, hence I hope there is no serious problem intersting behaviour missing -parallel... bye, francesco