CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   OpenFOAM (https://www.cfd-online.com/Forums/openfoam/)
-   -   solved issue of running out of virtual memory crashes (https://www.cfd-online.com/Forums/openfoam/99863-solved-issue-running-out-virtual-memory-crashes.html)

mihaipruna April 13, 2012 16:00

solved issue of running out of virtual memory crashes
 
I looked for this particular issue and nobody seems to have answered it.
The dreaded message: "This does not necessarily mean you have run out of virtual memory"...
Basically my SHM mesher would crash for large numbers of cells.
This seems to have fixed it:

On Ubuntu, Use this tutorial to increase the size of the swap partition or add a new one

https://help.ubuntu.com/community/SwapFaq

colinB April 17, 2012 07:22

Hi mihaipruna,

using swap partitions is basically assigning a part of the hard-drive as
RAM which the computer also may use to deal with big files.

But this is not an optimal solution to deal with the problem, since
the calculations (both mesh creation and solving) are slowed down
significantly.
This is due to the extra 'way' of communicating.
Without swap just RAM and Processor have to communicate and these
two are built that way to handle the amount of data the RAM can store
within a short time span.
Now also introducing a swap partition to increase the RAM capacities
this additional way of communication , which is actually not
meant to do so, but for storing (big) files permanently, slows down the
process, for the data has to be hauled forth and back from there as well.

I would be interested if the solver is actually not complaining about such a
big mesh or if he is also using the swap partition without further notes?

Another note:

In a different thread where you posted the link to this thread
I mentioned a rule of thumb for mesh creation with 1M cells per GB RAM.
This rule of thumb should be extended:
Don't use more than 2 GB of RAM per processor.
So the number of processors is somehow also limiting your mesh size.

In reality this means with a quadcore machine I'm limited to 8M cells as
long as 8 GB RAM are provided.

With these guidelines I got pretty good results without hitting any
hardware limits.They are however just proven for PCs not for clusters
or other 'super computers'

best regards

Colin

mihaipruna April 17, 2012 09:45

I'm not sure how OpenFOAM handles multiple processors, do you have to tell it to run in parallel mode?

wyldckat April 17, 2012 15:45

Greetings to all!

@Mihai:
Quote:

Originally Posted by mihaipruna (Post 355161)
I'm not sure how OpenFOAM handles multiple processors, do you have to tell it to run in parallel mode?

See the tutorial "./incompressible/simpleFoam/turbineSiting" as an example for running snappyHexMesh in parallel as well as running simpleFoam in parallel.
The "damBreak" tutorial (the 2nd one, I think) in the OpenFOAM User Guide explains the first steps into running OpenFOAM applications in parallel.

Best regards,
Bruno


All times are GMT -4. The time now is 07:07.