[roncresswell]

Hi,

I've got the pitzDaily example running in serial and because it's going to take a long time to run I wanted to look at how much speedup I get in parallel.

I ran into the same problems as mentioned in this thread:
Parallel processing September 08 - 07:03 am

and fixed them the same way. I changed "vectorSpace" into "fixedList" and, in my case, added a "value" entry to one of the inlet boundary conditions in the "U" file in the 0 directory before decomposing. That all went well and no errors reported.

I then started to run the model over 15 machines and the problem seemed to diverge after about 75 time steps, giving the following message:

--> FOAM FATAL ERROR : attempt to use janafThermo<equationofstate> out of temperature range 200 -> 5000;T = nan

From function janafThermo<equationofstate>::checkT(const scalar T) const
in file /home/dm2/henry/OpenFOAM/OpenFOAM-1.2/src/thermophysicalModels/specie/lnInclude/ janafThermoI.H at line 73.

FOAM parallel run aborting

So this indicates that the temperature has blown up - but the serial run is fine. The mesh was decomposed with metis by the way (oh, that's obvious from the number of processors isn't it?!). In fact the serial run is currently at about time step 7300.

Is this just numerics associated with decomposing into "too small" subdomains, or could there be an error introduced by the decomposition that wasn't picked up at the time?

[roncresswell]

Tried the 3D example instead - got a signal 11 on one of the nodes. Took that node out of cluster and redecomposed - running fine. Faulty memory on a node - therefore the above *might* indicate a bug, but most likely doesn't. Ignore.