If one is running Firefly in parallel, there is usually no need
to set mklnp or np in the input file. The exceptions are some
special types of jobs but certainly not a plain HF or DFT.
I'd suggest to run this job in parallel in the direct SCF mode
using a single scratch folder. I'd also suggest to leave mklnp, np,
httfix, httalt and httpar at their default values. I'd first try
to run this job using only 8 logical cores. I'd also suggest to run
it with 16 logical cores just to check whether this speeds things up or not.
For the run with 16 cores, it is necessary to add
$smp httfix=.f. $end
option to the input. Otherwise, it should be identical to the first one.
Hope this helps.
On Fri Jun 7 '13 1:39pm, Fumihito Mohri wrote
>Dear Prof. Alex Granovsky
>Thank you for your reply.
>>was it a parallel job? or just a multithreaded one?
>Yes, it is a parallel job with 16 partitions in only one HD ( c:\node1 c:\node2 ....c:\node16.)
>Furthermore, it is also a multithreaded job with two octa core + HTT.
>I think that the parallel job with only one HD is cause of the very slow computation. But, 16 HDs can not be mounted. So, choice of single + multithreaded job is better. Is this correct?
>Is the following condition appropriate for the single + multithreaded job?
> $SYSTEM MKLNP=16 $END
> $SMP HTTNP=2 $END
> $SMP httfix=.f. $END
> $SMP httpar=.t. $END
> $SMP httalt=.t. $END
> $SMP CSMTX=.t. $END
> $P2P P2P=.t. DLB=.t. $END
>>What was the calculation type?
>It is a usual geometry optimization using:
>$CONTRL ICHARG=0 MULT=1 SCFTYP=RHF RUNTYP=OPTIMIZE
> COORD=ZMTMPC MAXIT=200 NZVAR=0 DFTTYP=B3LYP $END
>$SYSTEM TIMLIM=600000 MWORDS=10 MKLNP=16 $END