|
[Sponsors] |
Parallel runs unstabel and composed stable, what is goinig on? |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
December 10, 2010, 04:58 |
Parallel runs unstabel and composed stable, what is goinig on?
|
#1 |
Senior Member
Claus Meister
Join Date: Aug 2009
Location: Wiesbaden, Germany
Posts: 241
Rep Power: 17 |
Hello Foamers,
has anybody ever made the following experience: a parallelized running the simulation breaks down. After reconstruction and letting the simulation run as a single process the error is disappeared and the simulation iterate over the problem time step. Afterwards, I decompose the case again and moves on. So, what happened? I though if I let the sim run in parallel or not, it doesn't matters the results or the behavior. Cheers in advance |
|
December 10, 2010, 05:09 |
|
#2 |
Senior Member
Bernhard
Join Date: Sep 2009
Location: Delft
Posts: 790
Rep Power: 21 |
Can you maybe post a case as simple as possible to reconstruct this behaviour? Some experienced users might be able to help you then. Maybe you can give some error messages?
|
|
December 10, 2010, 05:33 |
|
#3 |
Senior Member
Claus Meister
Join Date: Aug 2009
Location: Wiesbaden, Germany
Posts: 241
Rep Power: 17 |
Hello Bernhard,
unfortunately, I do not have a simple case and I cannot make up a simulation which exhibits a similar behavior. I discovered it accidentally. Error code: I don't have more as the usual error in OpenFOAM when a blow up has taken place. I think it could has something to do with the matrices which are computed on different cpus. In composed fashion, the might computed more easily. Have you ever made such experience? |
|
December 10, 2010, 06:12 |
|
#4 |
New Member
Denis Semyonov
Join Date: Mar 2009
Posts: 14
Rep Power: 17 |
I've had this kind of problem when was trying to use too many processors or some experimental domain decomposition (like decomposition of a simple domain along only one direction). To solve this problem it was just enough to decrease a number of the processors (not necessary to go to a single-processor mode).
As the speedup in that cases was anyway too low and there was no yield of using more processors I just forgot about that. |
|
December 14, 2010, 05:59 |
|
#5 |
Senior Member
Claus Meister
Join Date: Aug 2009
Location: Wiesbaden, Germany
Posts: 241
Rep Power: 17 |
That is usually what I do too; decreasing the number of processors, passing the critical position and then increasing the number of processor.
|
|
|
|