[aspect-devel] Number of grouped files = 1 -> code doesn't exit after it's finished

Rene Gassmoeller rene.gassmoeller at mailbox.org
Mon Mar 21 13:36:14 PDT 2016


Hi Jacky,
did you use this option before and did it work before? (and did maybe
something in your cluster evironment change?) Because if it changed its
behavior I might need to take another look at the recent changes. In
fact the behavior of this part  (when number of grouped files != 0)
should not have changed.

But it really seems that the MPIIO option is somehow implementation
dependent or at least fragile. E.g. on one of the clusters that I use I
recently noticed that I can set 'Number of grouped files = 24' and it
works flawlessly, but 'Number of grouped files = 32' does crash every
time (I am using more than 32 processes). It seems there is a limit for
concurrently active MPI communicators on this cluster. All in all I
think the MPIIO feature is very nice and useful for large models, but it
seems we can not guarantee its functionality on all systems.

If you get any insights from talking to your IT administrators, please
let us know. The relevant lines of code to look at would be
postprocess/visualization.cc:460-511 and
deal.II/source/base/data_out_base.cc::6137-6198.

Hope you can figure out the problem,
Rene



On 03/19/2016 11:06 AM, Austermann, Jacqueline wrote:
> Hi,
>
> After updating my master with the upstream master I’ve been running into the following problem: When I set Number of grouped files in subsection Postprocess, subsection Visualization to 1 (and probably to anything other than 0) the code runs through without problems till the end (after it prints the table with the different run times) but does then not exit. It just hangs and I can cancel it manually with the right output but if I don’t do that it just stays there until the runtime for the submitted run is exceeded. I tried this out with a few cookbooks and always had the same problem (I attached a .prm that produces this error).
> Let me know if you have any ideas why this could be.
>
> Thanks!
> Jacky
>
>
>
>
>
> _______________________________________________
> 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/20160321/88b7387f/attachment.html>


More information about the Aspect-devel mailing list