Writing parallel code

hartinger · August 9, 2005, 12:41

Hi,

I've written my own mesh deformation routine and want to run it in parallel, but doesn't work yet. I have no experience in that either. Could you give me a hint, where to start. And probably list some DOs and DONTs.

Thank you
Markus

mattijs · August 9, 2005, 14:07

Since you only are moving points (and not changing topology) you only have to make sure the point positions on processor patches are synchronized.

This is all done for you already in the motionSolver (see e.g. icoFoamAutoMotion.C) which determines internal point positions given boundary conditions.

If you want to do it yourself:
a. there are points on the processor patches shared between two processors only.
b. there are points shared between more than two processors.

a. is handled by sending one half to the other half and averaging so that both halves have the same value

b. is usually handled by sending all these coordinates to the master which determines a single value for all and sends it back.

Use motionSolver is my advice ;-)

hartinger · August 9, 2005, 14:20

thanks, i have a look on motionSolver
markus

skherad · March 11, 2006, 07:30

Dear all
is it possible to run engine cases(geometry deformation ) in parallel?
thank you for your time

mattijs · March 13, 2006, 13:07

The geometry deformation runs in parallel. There are some parallellizations problems though with other aspects of engine cases (notably injector specification). Have a search to find that one.

August 9, 2005, 12:41	Hi, I've written my own me	#1
hartinger Senior Member Markus Hartinger Join Date: Mar 2009 Posts: 102 Rep Power: 17	Hi, I've written my own mesh deformation routine and want to run it in parallel, but doesn't work yet. I have no experience in that either. Could you give me a hint, where to start. And probably list some DOs and DONTs. Thank you Markus

August 9, 2005, 14:07	Since you only are moving poin	#2
mattijs Senior Member Mattijs Janssens Join Date: Mar 2009 Posts: 1,419 Rep Power: 26	Since you only are moving points (and not changing topology) you only have to make sure the point positions on processor patches are synchronized. This is all done for you already in the motionSolver (see e.g. icoFoamAutoMotion.C) which determines internal point positions given boundary conditions. If you want to do it yourself: a. there are points on the processor patches shared between two processors only. b. there are points shared between more than two processors. a. is handled by sending one half to the other half and averaging so that both halves have the same value b. is usually handled by sending all these coordinates to the master which determines a single value for all and sends it back. Use motionSolver is my advice ;-)

August 9, 2005, 14:20	thanks, i have a look on motio	#3
hartinger Senior Member Markus Hartinger Join Date: Mar 2009 Posts: 102 Rep Power: 17	thanks, i have a look on motionSolver markus

March 11, 2006, 07:30	Dear all is it possible to ru	#4
skherad New Member saeid kheradmand Join Date: Mar 2009 Posts: 13 Rep Power: 17	Dear all is it possible to run engine cases(geometry deformation ) in parallel? thank you for your time

March 13, 2006, 13:07	The geometry deformation runs	#5
mattijs Senior Member Mattijs Janssens Join Date: Mar 2009 Posts: 1,419 Rep Power: 26	The geometry deformation runs in parallel. There are some parallellizations problems though with other aspects of engine cases (notably injector specification). Have a search to find that one.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Programmerfriendly code writing how do you do that	kar	OpenFOAM Running, Solving & CFD	5	January 25, 2008 11:33
Writing a CFD code	Ernesto	FLUENT	2	March 27, 2007 17:38
writing a batch for a parallel run	M	CFX	4	December 15, 2005 00:09
writing a cfd code	kelvin	Main CFD Forum	13	October 13, 2005 05:41
Code writing	Frank Scheuerpflug	Main CFD Forum	15	May 19, 2005 06:02