[Pw_forum] wfc files: heavy I/O, handling for restarts

Paolo Giannozzi giannozz at democritos.it
Mon Sep 5 19:08:54 CEST 2011


On Sep 5, 2011, at 5:22 , Michael Sternberg wrote:

> If one MPI rank writes a file, must this file indeed be seen or even
> readable by another MPI rank?

no, as long as there is no need to restart from previous interrupted
runs or to read wavefunction files for post-processing.

The amount of I/O is determined by options "wf_collect" and "disk_io".
The latter is also influenced by the number of k-points per process.
Note however that this holds only for pw.x; for other executables
(e.g. the phonon code) there are no such tricks. In all cases, it
should be possible to use local scratch directories with no problem,
collect files afterwards, if so desired. In principle, it shouldn't even
be needed to collect all files so that they are visibile to all  
processors,
but then you have to convince mpi to send the same process to the
same processor as in the previous run! It will however be impossible
to restart with a different number (or geometry) of processors.


Paolo
---
Paolo Giannozzi, Dept of Chemistry&Physics&Environment,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222







More information about the users mailing list