'Segmentation fault: 11' on running application in parallel

ulfaslak · November 27, 2013, 04:14

Dear foamers,

I'm having big trouble with 'Segmentation fault: 11', and I'd like to know if anyone else is experiencing something similar. I've searched the forum for answers but came up empty handed so here goes:

In a nutshell: I can run my case decomposed into 8 parts, but get nasty segfaults when I try with 16 parts. Decomposition is done using the 'simple'-method. The attached runLogs show output of both the 8-procs-run and the 16-procs-run, with the included error printed by terminal.

Further info: My jobs are sent to a high performance computing center, where the vast majority of nodes has 8 processors. I thought this could have something to do with it, as decomposePar needs hosts names of all the machines that the case is distributed to if it is not run one multiprocessor machine, but what confuses me is that I've previously been able to run the exact same case on 16 processors, on that same computing center, without error.

My second thought was that decomposePar splits the mesh into even parts, cutting straight through intersecting cells, consequently creating tiny cells in the decomposed domains, which would make the Courant Number explode in those cells (Courant Number is inversely proportional to cell volume so it would essentially be like dividing by zero, hence the segfault). However, I do not expect this to be the case as inspection indicates that it preserves each cell volume.

I'm running OF 2.2.0.

Thank you,

Ulf Aslak

Attachments:
https://www.dropbox.com/s/wszvnmj4iowa60f/runLog8
https://www.dropbox.com/s/do1q0cimj19zx0m/runLog16
https://www.dropbox.com/s/sm7hop8568...composeParDict

Aadhavan · November 27, 2013, 05:28

Hi there,
Yes, it is problem with OF cutting algorithm. There is no exact solution to this problem, However what you can do is, play around decomposition methods (scotch, simple, hierarchical ....) then you may solve your problem.

Thanks,
Aadhavan

ulfaslak · November 27, 2013, 07:57

Dear Aadhavan,

Thanks for the quick response! I see you've had a similar problem (guess I didn't search the forum thoroughly enough).

How did you arrive at this conclusion? And do you have any tips and tricks on how to decompose successfully? I'm simulating on a pretty fine mesh, so a trail-and-error approach would take for ever, I'm afraid.

November 27, 2013, 04:14	'Segmentation fault: 11' on running application in parallel	#1
ulfaslak New Member Ulf Aslak Join Date: Sep 2013 Location: Copenhagen, Denmark Posts: 5 Rep Power: 12	Dear foamers, I'm having big trouble with 'Segmentation fault: 11', and I'd like to know if anyone else is experiencing something similar. I've searched the forum for answers but came up empty handed so here goes: In a nutshell: I can run my case decomposed into 8 parts, but get nasty segfaults when I try with 16 parts. Decomposition is done using the 'simple'-method. The attached runLogs show output of both the 8-procs-run and the 16-procs-run, with the included error printed by terminal. Further info: My jobs are sent to a high performance computing center, where the vast majority of nodes has 8 processors. I thought this could have something to do with it, as decomposePar needs hosts names of all the machines that the case is distributed to if it is not run one multiprocessor machine, but what confuses me is that I've previously been able to run the exact same case on 16 processors, on that same computing center, without error. My second thought was that decomposePar splits the mesh into even parts, cutting straight through intersecting cells, consequently creating tiny cells in the decomposed domains, which would make the Courant Number explode in those cells (Courant Number is inversely proportional to cell volume so it would essentially be like dividing by zero, hence the segfault). However, I do not expect this to be the case as inspection indicates that it preserves each cell volume. I'm running OF 2.2.0. Thank you, Ulf Aslak Attachments: https://www.dropbox.com/s/wszvnmj4iowa60f/runLog8 https://www.dropbox.com/s/do1q0cimj19zx0m/runLog16 https://www.dropbox.com/s/sm7hop8568...composeParDict

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Segmentation fault running waveDyMFoam solver (mod. interDyMFoam solver - waves2Foam)	Ed R	OpenFOAM Running, Solving & CFD	5	July 2, 2013 11:36
segmentation fault when installing OF-2.1.1 on a cluster	Rebecca513	OpenFOAM Installation	9	July 31, 2012 15:06
Segmentation Fault	Shawn_A	OpenFOAM Running, Solving & CFD	6	October 31, 2011 14:38
Segmentation fault in running alternateSteadyReactingFoam,why?	NewKid	OpenFOAM	18	January 20, 2011 16:55
Segmentation fault with parallel computing	Prad	Main CFD Forum	0	December 11, 2008 07:44

November 27, 2013, 05:28		#2
Aadhavan Member Aathavan Join Date: Nov 2012 Posts: 70 Rep Power: 13	Hi there, Yes, it is problem with OF cutting algorithm. There is no exact solution to this problem, However what you can do is, play around decomposition methods (scotch, simple, hierarchical ....) then you may solve your problem. Thanks, Aadhavan

November 27, 2013, 07:57		#3
ulfaslak New Member Ulf Aslak Join Date: Sep 2013 Location: Copenhagen, Denmark Posts: 5 Rep Power: 12	Dear Aadhavan, Thanks for the quick response! I see you've had a similar problem (guess I didn't search the forum thoroughly enough). How did you arrive at this conclusion? And do you have any tips and tricks on how to decompose successfully? I'm simulating on a pretty fine mesh, so a trail-and-error approach would take for ever, I'm afraid.