[Wien] Speedup of heavy cases?

[Peter Blaha]

> I have some $cases that produce multi-GB vector files, and the time
> spent in I/O in lapwso and lapw2c are becoming a problem, and the
> problem only gets worse if I use k-point parallelization on an
> 8-processor machine. As lapw1 only writes the vector files, it does not
> seem to cause any trouble with the I/O.
>
> Will it help me if I do a simple rewrite of runsp_lapw to support 2
> scratch partitions, and have the two partitions on two physically
> different disks? I have in mind, that lapwso could read from the usual
> $SCRATCH and write to another partition, say $SCRTWO, and that lapw2c
> does it the opposite way around afterwards (the help* files), or are
> there more complications to expect from this?

Are these $SCRATCH spaces "local" disks (not NFS mounted!) ? This is
vital for good performance of large vector files. If it is via NFS, good
NFS options are necessary, otherwise it may slow down to 20%, but with a
good NFS system even NFS writing should not be a problem.

I'm not sure that 2 scratch spaces really help, but you may try.
Modifications are necessary in x_lapw (not runsp) and should be quite simple.

PS: Be sure that case.inso has a as small as possible E-window ! This
should limit the size of the vectorso files (lapw1 probably needs a large
E-range for larger variational freedom, but in lapwso you should write
ONLY the required E-range)


                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------