Dan,

I found the problem. I was assuming I had a global value in a subroutine but 
only had a local value (a sum). 

Now running correctly across multiple nodes and cores!

Thanks,

Bill

On Apr 23, 2014, at 10:15 AM, Daniel Wheeler <[email protected]> wrote:

> On Tue, Apr 22, 2014 at 3:34 PM, Seufzer, William J. (LARC-D307)
> <[email protected]> wrote:
>> FiPy,
>> 
>> I have been successful with getting Trilinos installed and running on our 
>> cluster. Thanks for helping to resolve the parallelization issues but now 
>> there's another matter.
>> 
>> I have a volumetric heat source and if I add 50 Joules of energy into a mesh 
>> during a short run, then at the end I have been able to use Q= mass * 
>> heatCapacity * deltaTemperature across the mesh and I find 50 Joules. I'm 
>> using constant density and heatCapacity values (no Temperature dependance).
>> 
>> I'm confident I have the heat source and FiPy equation set up correctly.
>> 
>> However, when I run in parallel I find more that 2x the amount of energy and 
>> higher temperatures in the mesh. I've run 3 different mesh densities and the 
>> problem changes with mesh density but appears independent of the step size 
>> (delta Time).
>> 
> 
> A few things you might want to check:
> 
>  * if you are calculating a global value, make sure it is actually
> being evaluated correctly and you are not just getting the local
> (global) value
> 
>  * if you are using the global value as part of the calculation and
> not just a post-process then it might lead to the wrong field values
> if it is calculated incorrectly
> 
>  * check that the solver is converging in parallel
> 
> -- 
> Daniel Wheeler
> 
> _______________________________________________
> fipy mailing list
> [email protected]
> http://www.ctcms.nist.gov/fipy
>  [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]


_______________________________________________
fipy mailing list
[email protected]
http://www.ctcms.nist.gov/fipy
  [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]

Reply via email to