[Pw_forum] ph.x high I/O frequency

Paolo Giannozzi giannozz at democritos.it
Wed Oct 17 08:45:37 CEST 2012

On Oct 17, 2012, at 5:05 , Jiseok Kim wrote:

> So I tried to reduce the I/O by setting the option 'reduce_io=.true.'.

I don't think it has much effect on the phonon code. It only affects
the first step, the self-consistent calculation.

The current version of the phonon code follows an old paradigm:
all large arrays depending upon the k point (Kohn-Sham orbitals
\psi_k, perturbations \Delta V\psi_k, linear responses \Delta\psi_k)
are computed and stored to file. Only a single k-point is kept in
memory. This behavior can be changed only by modifying the code.

One can reduce the amount of load on the system by writing to a
local filesystems instead of a parallel file system. I am not sure it
works in the phonon case, though. In any case, the files end up
scattered among different processors so there are many problems
and limitation (no possibility to restart, for sure).

In conclusion: there are many possible ways to improve the behavior
of the phonon code with respect to I.O load, but none is simple, most
require some serious work.

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