CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM Announcements from Other Sources (http://www.cfd-online.com/Forums/openfoam-news-announcements-other/)
-   -   multiSolver released (http://www.cfd-online.com/Forums/openfoam-news-announcements-other/78502-multisolver-released.html)

marupio July 23, 2010 08:30

multiSolver released
 
Hi Foamers,

I am pleased to announce the release of multiSolver, an extension to OpenFOAM that allows you to run numerous solvers in sequence on a single dataset.

Why would you need that?

  • Arbitrarily changing boundary conditions mid-run;
  • Modelling multi-step processes in a single application (e.g. injection & settling);
  • Modelling a problem characterized by 2 different timesteps (e.g. stirring & biochemical reactions).
See this page on the wiki for further details:

http://openfoamwiki.net/index.php/Contrib_multiSolver

Share and enjoy!

chegdan March 28, 2011 16:04

Parallelization?
 
Are parallel runs supported yet? I'm using OF-1.6-ext and this would be fantastic for my applications.

Dan

marupio March 28, 2011 17:20

Hi Dan,

No, it isn't. I haven't worked with parallel very much, so I'm not familiar with the directory structure and how the processors communicate. I don't forsee it being challenging, but I don't have the time right now. I would be open to collaborating on its development, though!

I expect to be putting mesh motion in this summer. I'll also have an updated version with minor changes before the 6th workshop.

-Dave

chegdan March 28, 2011 17:36

Quote:

Originally Posted by marupio (Post 301272)
Hi Dan,

No, it isn't. I haven't worked with parallel very much, so I'm not familiar with the directory structure and how the processors communicate. I don't forsee it being challenging, but I don't have the time right now. I would be open to collaborating on its development, though!

I expect to be putting mesh motion in this summer. I'll also have an updated version with minor changes before the 6th workshop.

-Dave


I would be willing to collaborate on this one to get it working. I will take a look at the code and see if I can offer any help. It looks like a fantastic tool and could definitely be worth getting it to work in parallel. I'll get into it starting tomorrow.

Dan

marupio March 28, 2011 21:08

Grand! I've sent you a PM with some more information. I hope we can get this working!

marupio April 6, 2011 10:19

multiSolver now fully supports parallel simulations.

canfield May 20, 2011 06:19

Hello. First of all, great work with this solver!
I do seem to have hit an obstacle though. You say it supports parallel processing. I don't seem to be able to figure out how to get the solver to run. I have decomposed the case as explained on the WIKI page, but when I try to run the case using mpirun .... it won't run. It seems that there are no "0" directories in the processor directories. I guess I am just missing something trivial, but I haven't worked it our yet. Maybe you can give me a tip?

marupio May 20, 2011 09:17

It sounds like you aren't using a multiSolver-enabled solver. What solver are you using? The only publicly available one is multiSolverDemo, which came with the installation. The wiki page also shows how to easily create a multiSolver-enabled application.

If you are using the correct solver... then what errors are you getting?

canfield May 20, 2011 09:26

OK, here goes. I compiled my own multisolver called multiBubbleSolver. Incidentally, it works fine on a single processor. Then I decompose as per instructions on the wiki. then I input:


mpirun -np 4 multiBubbleSolver -parallel


and I receive this:


[1] --> FOAM FATAL ERROR:
[1] Initial time directory missing for solver domain [interFoam1]. Expecting "/home/peter/OpenFOAM/peter-1.7.x/run/BUBBLES/processor1/multiSolver/interFoam1/initial/0"
[1]
[1] From function multiSolver::checkTimeDirectories
[1] in file multiSolver/multiSolver.C at line 320.

...The rest of the errors seem to be caused by the missing directory.

The same thing happens if I try using multiSolverDemo in parallel.The directory structure in the proc* folders is as follows ("0" directory missing):

processor*
-constant
-system

marupio May 20, 2011 09:46

Hi Peter, I've managed to duplicate the issue on my machine. I'll try to resolve it as quickly as I can, but I have to leave for work soon. I'll let you know...
-Dave

marupio May 20, 2011 10:29

Hi Peter, I've got a work-around you can try for now... the bug fix should come within a few days. Edit the controlDict file (not the multiControlDict), and set:

startFrom: startTime;
startTime: -1;

Then try the long decompose command in the wiki. I have to head off to work now. Hope it works!

canfield May 20, 2011 10:38

Nice job! works fine for the demo; I'll have a go at my case later on. Thanks.

marupio June 3, 2011 09:44

The bug fix has been completed. Parallel decomposing should nolonger omit the initial directories!


All times are GMT -4. The time now is 08:02.