Re: FLUKA: Re: FLUKA
Hi all
the procedure suggested by Stefan is sound and should work nicely. An
even better way of achieving the same result is to create n copies
of the input file, myinput.inp, myinputb.inp, myinputc.inp...
and then
cp ranskip001 ranmyinput001
cp ranskip002 ranmyinputb001
cp ranskip003 ranmyinputc001
and run in parallel
{bsub -q ..} rfluka -N0 -MX myinput
{bsub -q ..} rfluka -N0 -MY myinputb
{bsub -q ..} rfluka -N0 -MZ myinputc
where X,Y,Z are the number of runs you want to get from each machine.
The default number in rfluka if the -M option is not given is 5. This
allows to go on with potentially unlimited number of runs and no clash.
Assuming myinput has run for 5 times and you want to run for extra 3 ones
you can always issue:
{bsub -q ..} rfluka -N5 -M8 myinput
Please note -N is pointing to the number of the LAST executed run.
If somebody is interested in having widely space random numbers which can
be used as more or less independent starting seeds let me know. Stefan was
offering these for Linux, I have sets for Digital Unix and IBM-AIX as well
and maybe for HP-UX (I must check)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
VERY IMPORTANT
NEVER USE RANDOM NUMBER SEEDS GENERATED ON A GIVEN ARCHITECTURE
ON A DIFFERENT ONE!!! THEY WILL NOT WORK (THE FILE IS IN HEXADECIMAL
FORMAT AND ACTUALLY DEPENDS ON THE SYSTEM) AND IF YOU ARE LUCKY
YOU WILL RUN INTO A CRASH, IF YOU ARE UNLUCKY THE CODE COULD KEEP
RUNNING BUT THE RANDOM NUMBER GENERATOR COULD BE NOT RANDOM
AT ALL WITH OBVIOUS CONSEQUENCES (IT OCCURRED IN THE PAST....
ALL NUMBERS WERE COMING OUT BETWEEN 0 AND 1E-5)
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Ciao
Alfredo
On Thu, 29 Mar 2001, Stefan Roesler wrote:
> Hi Laura,
>
> You can can in parallel if you start each job with start seeds which are
> sufficiently spaced in the sequence of random numbers. For Linux we have
> collected at SLAC a set of 17 of such start seeds which you find in the
> attached package (ranskip001,...,ranskip017, where ranskip001 is the default
> random.dat).
>
> What I normally do is the following: Assuming your input-file is called
> myinput.inp, copy
>
> cp ranskip001 ranmyinput001
> cp ranskip002 ranmyinput011
> cp ranskip003 ranmyinput021
> ...
>
> and then start in parallel
>
> {bsub -q ..} rfluka -N0 -M1 myinput
> {bsub -q ..} rfluka -N10 -M11 myinput
> {bsub -q ..} rfluka -N20 -M21 myinput
> ...
>
> After this set of runs is finished you can simply resubmit another set by
>
> {bsub -q ..} rfluka -N1 -M2 myinput
> {bsub -q ..} rfluka -N11 -M12 myinput
> {bsub -q ..} rfluka -N21 -M22 myinput
> ...
>
> which would use as start seeds those created at the end of the first set
> of runs, etc.
>
> Regards,
> Stefan
>
>
> On Thu, 29 Mar 2001, Laura Baudis wrote:
>
> > Hi Stefan,
> >
> > I have again a short question about FLUKA.
> > If I want to run the same job simultaneously on different
> > machines I need to initilize the random seeds differently for each of the
> > jobs. Is this possible or do I have to run a job first and take the
> > seeds which are written at the end of the run?
> >
> > Regards,
> > Laura
> >
>
>
> ___________________________________
>
> Stefan Roesler
> SLAC
> P.O. Box 4349, MS 48
> Stanford, CA 94309
>
> Tel: 650-926-2048
> Fax: 650-926-3569
> E-mail: sroesler@slac.stanford.edu
> Stefan.Roesler@cern.ch
>
>
--
+----------------------------------------------------------------------------+
| Alfredo Ferrari || Tel.: +41.22.767.6119 |
| CERN-SL || Fax.: +41.22.767.7555 |
| 1211 Geneva 23 || e-mail: Alfredo.Ferrari@cern.ch |
| Switzerland || Alfredo.Ferrari@mi.infn.it |
+----------------------------------------------------------------------------+
Partial thread listing: