[aspect-devel] A weird error with MPI postprocessing

FELIPE ORELLANA ROVIROSA f_orellana at berkeley.edu
Mon Feb 29 14:17:15 PST 2016


 Hi Wolfgang,

    Thanks a lot for replying.

    I am running simple convection in a brick with lengths (2,1,1). I am
just using the 3D-convection example from the cookbook and simply modifying
it to be a brick and not a cube. I have succeeded, slowly, at Ra10^6.
Nevertheless, only some of the output files are readable.

    I reached a maximum time of t=0.00450949 seconds  (with a max time =1)


    Now I am trying to modify that brick, with a different initial
condition on temperature, putting instead a 3D Gaussian, at the brick's
center, with amplitude of 0.01 or 0.02, and radial length scale of 0.25
(dies-off well inside the brick).

    I have not been able to get readable .pvtu files, with error messages
like:

           'File for piece xy cannot be read'           (with xy being a
cell number)

  I can only load some vtu parts of the domain..

   So, the amplitude of the initial condition for temperature is small
enough compared to the boundary ones (0 and 1), and is as small as the one
on the example, with the only substantial difference being that mine one is
''monopole'', it's just hot (or just cold) (unlike the example one, which
is dipole-type), so that may contradict the global Rayleigh number a bit,
at least initially. I attach the parameter file. I hope the refinement
parameters (initial global, local, and time-refinement) can handle this
owing to the similarity with the example.

      I think my cluster is not stable enough, I am running the task on 8
nodes. But lots of failures, sometimes segmentation faults, at random times.

hope you can understand,

cheers,
thanks,

Felipe

On Mon, Feb 29, 2016 at 1:47 PM, Wolfgang Bangerth <bangerth at tamu.edu>
wrote:

>
> Felipe,
>
>      On average, what percentage of the Aspect output files (pvtu for
>> example)
>>   can you actually and successfully read in Paraview? I mean with no
>> errors.
>>
>> I know errors can come from the cluster one is using, but also from bugs
>> in
>> the software; so it's hard to keep track, and also hard and tedious to
>> advance..
>>
>>      I am honestly feeling a little bit discouraged on all this effort I
>> have
>> put in Aspect. It's just that I can hardly advance on my research., I have
>> tons of models I wanna run, but I hardly succeed..
>>
>
> What concretely are the problems you are seeing?
>
> The thing is that the majority of us regularly do very large computations,
> get lots of VTU files, and visualize them successfully. See for example of
> the videos posted here:
>   http://aspect.dealii.org/gallery.html
> Most of them were run on hundreds to a few thousands of cores.
>
> Of course that doesn't mean that *all* of us are lucky enough to have
> clusters that work right away, but most of us seem to, and most of the rest
> have eventually gotten their runs to work ultimately.
>
> Best
>  Wolfgang
>
> --
> ------------------------------------------------------------------------
> Wolfgang Bangerth               email:            bangerth at math.tamu.edu
>                                 www: http://www.math.tamu.edu/~bangerth/
>
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20160229/6db2ae51/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: r3Dbic.prm
Type: application/octet-stream
Size: 6726 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20160229/6db2ae51/attachment-0001.obj>


More information about the Aspect-devel mailing list