[Pw_forum] Intensive load of cpu, mpich during pw.x or not?

Paolo Giannozzi giannozz at nest.sns.it
Mon Apr 18 20:01:56 CEST 2005


On Monday 11 April 2005 14:49, Sergey Lisenkov wrote:

> I wondering how frequently pw.x writes/reads data to/from the disks

quite frequently, since it was designed to minimize the amount 
of things kept in memory:

- In the initialization phase, the potential, the indices of k+G and
wavefunctions for all k are written. 

- For each scf iteration and for each k-point, one reads [*] the 
indices of k+G and the wavefunctions from previous scf iteration, 
and writes the new wavefunctions.

- When the new charge density is calculated, no I/O is required
unless disk_io='high'; then the new potential is calculated and
written to file. 

- At each structural optimization step there is some additional I/O 
for charge and wavefunctions extrapolation.

[*] unless there is only one k-point per processor, in which case
there is no point in re-reading what is already in memory

It shouldn't be difficult to keep everything in memory, with relatively
little disruption to the code, but it is something that must be done
with great care. 

> how intensively it uses mpi?

a lot. If the R- and G-space grids are distributed, each 3d FFT uses 
mpi.

P.

-- 
Paolo Giannozzi             e-mail:  giannozz at nest.sns.it
Scuola Normale Superiore    Phone:   +39/050-509876, Fax:-563513 
Piazza dei Cavalieri 7      I-56126 Pisa, Italy



More information about the Pw_forum mailing list