[aspect-devel] aspect errors on TACC stampede

Timo Heister heister at clemson.edu
Wed Mar 16 02:48:43 PDT 2016


Hey Robert,

I haven't seen this issue before. The random nature makes me wonder if
this is a problem of the system. Do you see any other pattern with the
failures, for example the node in question is always the same?

On Tue, Mar 15, 2016 at 8:49 PM, Robert Martin-Short
<rmartin-short at berkeley.edu> wrote:
> Dear Aspect development team
>
> I'm running Aspect 1.3 with the latest version of deal.ii and trilinos
> 11.12.1 on the TACC stampede cluster. I've been finding that my simulations
> often fail after several timesteps with the error
>
>  Solving Stokes system... aspect:
> /tmp/trilinos-11.12.1-Source/packages/epetra/src/Epetra_BasicDirectory.cpp:367:
> int Epetra_BasicDirectory::Generate(const Epetra_BlockMap&) [with int_type =
> int]: Assertion `curr_LID !=-1' failed.
> [c559-001.stampede.tacc.utexas.edu:mpi_rank_16][error_sighandler] Caught
> error: Aborted (signal 6)
>
> This behavior is very inconsistent - sometimes a simulation will run without
> error, then I'll start it again and it will fail with this error after a few
> minutes. I am running setups that are known to work on another cluster, so
> I'm wondering if this problem has something to do with the combination of
> aspect/deal.ii/trilinos that I'm using?
>
> Can someone help me understand what this means and how to fix it?
>
> Thanks very much
>
> Robert
> --
> Robert Martin-Short
> Graduate Student
> Department of Earth and Planetary Science
> U.C Berkeley
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel



-- 
Timo Heister
http://www.math.clemson.edu/~heister/


More information about the Aspect-devel mailing list