[Pw_forum] espresso 2.1.4 and ifc8.1.024

[Axel Kohlmeyer]

On Fri, 1 Jul 2005, Patricia Paredes wrote:

patricia,

PP> I use the same input for the old and new pwscf.

i have run your input with a serial 32-bit binary on an
opteron 246 machine and i need 42.85 seconds for the
first scf iteration and 403 seconds for the first
ionic step. so i guess it cannot be the intel compiler.

PP> There are some differences in the energies through the 
PP> scf iterations, in the last 4 digits,

PP> as a result of these small diferences, it took more ionic iterations
PP> to converge for the expresso version of the code.

i did not follow all the details, but there have been changes
and corrections to several parts of the code that affect
the total energies and also the way the potential and
wavefunctions are extrapolated, between the ionic steps since
version 2.0. 
in some cases, such 'improvements' can result in requiring
a few more steps in the optimization and differences in
the final total energies. it is not always easy to tell
whether this is significant or not. 

PP> It was running on a single node, using the parallel code. Note that it
PP> was a Gamma point calculation.

[...]
PP> >do you have hyper-threading enabled or disabled?
PP> 
PP> It is enabled

this is a point you should watch. for numerical codes
this is actually harming performance more often than
helping. that would, however, be totally independent
from the compiler and the pwscf version.

PP> >are you running in parallel or serial?

PP> parallel version, but only in one node (since it was 
PP> a Gamma point input)

you don't need k-points for running in parallel. 
pw.x also supports distributed memory parallelization
of the G and R space. this is less efficient than
parallelizing across k-points but with a fast network
like myrinet or infiniband it should even scale 
reasonably well up to quite a few nodes.

PP> >the same communication library and hardware?
PP> 
PP> 
PP> Yes!!

ok, you may have a problem, if the MPI was compiled
with ifc7 and you compile espresso with ifort8.1

for that you may want to test compiling pwscf2.0 with
ifort8.1 to see.

PP> >what error?
PP> 
PP> p4_10080: (24418.614286) net_recv failed for fd = 8
PP> 
PP> p4_10080:  p4_error: net_recv read, errno = : 104

ouch. that does not look too good. there is something
fishy going on. perhaps you have a bad network 
connection (and thus a lot of re-transmits) somewhere.

i hope this will help you a little more to figure out 
what is going wrong. locating the problem when you change
several parameters simultaneously is never simple.

best regards,
	axel.

-- 
=======================================================================
Dr. Axel Kohlmeyer   e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.