CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > OpenFOAM Bugs

Surface Vector Field Problem at Parallel

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   June 29, 2010, 05:04
Default Surface Vector Field Problem at Parallel
  #1
Member
 
John Wang
Join Date: Mar 2009
Location: Singapore
Posts: 71
Rep Power: 8
cwang5 is on a distinguished road
Hi,

I'm running OF-1.6.x with a modified version of pimpleDyMFoam, and is using the function mesh.Sf().boundaryField() to calculate the patch area, which was later used to calculate the forces acting on the patch. The function was setup similarly to the force calculation in forces.C.

The surface vector field works fine in serial computation, but when I tried running it in parallel, under certain conditions the mesh.Sf().boundaryField() would return zero value, and the forces/forceCoeffs directory would not be created even though the functions were specified in the controlDict file. The problem seems to happen as long as the domain decomposition was not done along the spanwise direction (z-direction).

I'm not quite sure if it's a problem with the decomposePar code or the Sfb function itself, but would appreciated if someone can suggest a way around this.

John

EDIT: Correction, I've just realized that the forces directory did not show up because I stopped the simulation before the first output, I will update the behavior of the forces functions when I reach that point after restarting the simulation.
cwang5 is offline   Reply With Quote

Old   June 29, 2010, 06:58
Default
  #2
Super Moderator
 
Mattijs Janssens
Join Date: Mar 2009
Posts: 1,416
Rep Power: 16
mattijs is on a distinguished road
mesh.Sf().boundaryField() is per patch the area. I assume that you mean you get a zero size for some patches if they have zero faces on that particular processor - which is as should be.

Is there still a problem?
mattijs is offline   Reply With Quote

Old   June 29, 2010, 07:12
Default
  #3
Member
 
John Wang
Join Date: Mar 2009
Location: Singapore
Posts: 71
Rep Power: 8
cwang5 is on a distinguished road
The forces function itself works fine. I guess it's how I wrote my code that's causing the problem. You might want to move this post under the OF programming folder if that's the case.

Anyway, this is the code that I used to calculate the patch area:
Code:
scalar pArea = 0;

forAllConstIter(labelHashSet, patchList_, iter)
{
        label patchi = iter.key();
        dimensionedScalar patchArea = sumMag(Sfb[patchi]);
        pArea += patchArea.value();
}
the pArea is actually a sum of patchArea throughout, so it shouldn't matter even if the patch doesn't exist in one of the processor domains, right?

I suppose it has to do with my reading the patch list pointer in the main function under
Code:
const dynamicFvmesh& mesh = U.mesh();
instead of the
Code:
const fvMesh& mesh = refCast <const fvMesh>(obr_)
as specified in Foam::forces::read


EDIT: Apparently the cause of the problem is slicing the domains right at the patch in question.

Last edited by cwang5; June 29, 2010 at 10:50.
cwang5 is offline   Reply With Quote

Old   June 30, 2010, 05:05
Default
  #4
Super Moderator
 
Mattijs Janssens
Join Date: Mar 2009
Posts: 1,416
Rep Power: 16
mattijs is on a distinguished road
>Apparently the cause of the problem is slicing the domains right at the patch in question.

That should not be a problem. All processors will have all (non-processor) patches. Just that they might have zero faces in them.
mattijs is offline   Reply With Quote

Old   July 1, 2010, 03:47
Default
  #5
Member
 
John Wang
Join Date: Mar 2009
Location: Singapore
Posts: 71
Rep Power: 8
cwang5 is on a distinguished road
Quote:
Originally Posted by mattijs View Post
>Apparently the cause of the problem is slicing the domains right at the patch in question.

That should not be a problem. All processors will have all (non-processor) patches. Just that they might have zero faces in them.
Yes, I do see a patches presented in the processorX/constant folder, however, the the summation of sumMag(Sfb[patchi]) function results in zero area if the patch in question is not presented in all partitions.

The case that I am running is attached below to clarify the problem that I am encountering:
If the decomposition was done (using "simple") along the z-axis (xy-plane), the patch area will be computed correctly (although it appears that the motion between different portion of the divided patch doesn't seem to match exactly, I'm still working on that)
However, if the decomposition was done along the x or y axis (hence the patch 'flap' will not exist in one of the partitions) the patch area computed from the code stated in the previous post would return zero value.

Anyway, I am in the process of switching to 1.7 right now in hope that the dynamic mesh solver in the new version would somehow solve the problem. Thanks for taking the time to reply to my messages.

John

The case folder was too big to be uploaded to the forum, so I'm uploading it to rapidshare.
Attachment:
http://rapidshare.com/files/40422175...se.tar.gz.html
cwang5 is offline   Reply With Quote

Old   July 1, 2010, 05:41
Default
  #6
Super Moderator
 
Mattijs Janssens
Join Date: Mar 2009
Posts: 1,416
Rep Power: 16
mattijs is on a distinguished road
That makes it clear - use gSumMag (does additional global sum) instead of sumMag.
mattijs is offline   Reply With Quote

Old   July 12, 2010, 08:31
Default
  #7
Member
 
John Wang
Join Date: Mar 2009
Location: Singapore
Posts: 71
Rep Power: 8
cwang5 is on a distinguished road
Quote:
Originally Posted by mattijs View Post
That makes it clear - use gSumMag (does additional global sum) instead of sumMag.
Oh, wow... that did the trick, thank you so much.

John
cwang5 is offline   Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
problem of running parallel Fluent on linux cluster ivanbuz FLUENT 11 March 10, 2010 16:13
open channel problem (free surface) Andy Chen FLUENT 4 July 10, 2009 01:20
curve surface problem sianG CFX 0 February 8, 2009 09:52
parallel problem rui CD-adapco 2 July 31, 2007 13:23
CFX Free Surface problem sam CFX 6 March 21, 2007 19:49


All times are GMT -4. The time now is 18:12.