[aspect-devel] Fwd: error when writing checkpoint files

Timo Heister heister at clemson.edu
Thu Jul 5 14:32:38 PDT 2018


Magali,

> What is suppose to happen during checkpointing - is each processor writing its own files to the output directory?

The crash happens inside p4est that is writing the restart.mesh file
as far as I can tell from reading your error message. If I remember
correctly, this file is written from all processors and they do some
trickery that might fail on some network filesystem (even though I
have never seen it fail). That is why I suggested trying if there is a
different filesystem you can use (for example our cluster has a slow
NFS for /home/ but a fast network filesystem for IO).

> Is this handled differently than the visualization output, which for VTK output also writes individual files for each processor?

Yes, because the vtk output is either written in separate files per
processor or into larger files (see the "grouping" parameter) using
parallel MPI I/O. I wrote that code, though. ;-)

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


More information about the Aspect-devel mailing list