mpi raxml run on a cluster

[pavlos]

Hi raxml community,

I got a bit confused trying to run raxml on a linux cluster. For some reason, i decided to use the mpi version

the submission script looks like:
#SBATCH -J raxml

#SBATCH --clusters=mpp1

#SBATCH --get-user-env

#SBATCH --ntasks=64

#SBATCH --mail-type=end

#SBATCH --mail-user=j...@gmail.com

#SBATCH --export=NONE

#SBATCH --time=24:00:00

source /etc/profile.d/modules.sh

srun_ps /home/hpc/uj502/uj502kj/temp/raxml.sh

and the raxml.sh contains:
raxmlHPC-MPI-SSE3 -n bousios -s pol.complete.all.NameMod.mafft -N 64 -m GTRGAMMA -p 54321

My plan is to run 64 independent analyses, but it seems that i'm not doing that exactly because there have been produced about 200 trees already.

so what am I doing? :)

all the best
pavlos

[Alexandros Stamatakis]

Hi Pavlo,

That's interesting, since:

#ifdef _WAYNE_MPI
if(n % processes != 0)
n = processes * ((n / processes) + 1);
#endif

otherwise n = processes ... are you sure you got over 200 tree files
that all have the same type of name?

Alexis

> --
> You received this message because you are subscribed to the Google
> Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to raxml+un...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>

--
Alexandros (Alexis) Stamatakis

Research Group Leader, Heidelberg Institute for Theoretical Studies
Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology
Adjunct Professor, Dept. of Ecology and Evolutionary Biology, University
of Arizona at Tucson

www.exelixis-lab.org

[Pavlos Pavlidis]

Hi Alexi,

we checked it a bit with Pashalia a couple of hours ago. It seems that i was doing something stupid and thus I was probably wrong, but perhaps there is still some problem that I'd like to check a bit later.

best

pavlos

[David Kainer]

Hi Alex,

i'm also a little confused with the MPI version on a cluster. I ran the following job on a cluster:

#!/bin/csh
#PBS -wd
#PBS -q normal
#PBS -l walltime=00:10:00,vmem=192000MB,jobfs=400Mb,ncpus=64
module load ipm
module load openmpi
setenv TMPDIR $PBS_JOBFS
mpirun raxmlHPC-MPI-SSE3 -m GTRGAMMAI -N 20 -p 12345 -s alignment.phy -q PFaicc_GTRIG_toRaxml.txt -n PFaicc


Now, i set '-N 20' but set ncpus=64. I ended up with 64 trees. Is that expected?
If I want to end up with 20 trees should i have just selected ncpus=20?

cheers
David

[Tobias Klöpper]

Hi David,
as far as I know RAxML will run one calculation for each CPU and round
up to the number of CPUs you provide. So if you set -N 1000 but give the
process 2000 CPUs you will end up with 2000 trees.

I think it is just the easiest way to make this process parallel, as the
calculation for each tree is independent.

Cheers,

Tobias

> Hi Alex,
>
> i'm also a little confused with the MPI version on a cluster. I ran the
> following job on a cluster:
>
> #!/bin/csh
> #PBS -wd
> #PBS -q normal

> #PBS -l walltime=00:10:00,vmem=192000MB,jobfs=400Mb,*ncpus=64*

> module load ipm
> module load openmpi
> setenv TMPDIR $PBS_JOBFS

> mpirun raxmlHPC-MPI-SSE3 -m GTRGAMMAI*-N 20* -p 12345 -s alignment.phy

> > #SBATCH --mail-user=j...@gmail.com <javascript:>

> >
> > #SBATCH --export=NONE
> >
> > #SBATCH --time=24:00:00
> >
> > source /etc/profile.d/modules.sh
> >
> > srun_ps /home/hpc/uj502/uj502kj/temp/raxml.sh
> >
> > and the raxml.sh contains:
> > raxmlHPC-MPI-SSE3 -n bousios -s pol.complete.all.NameMod.mafft -N 64
> > -m GTRGAMMA -p 54321
> >
> > My plan is to run 64 independent analyses, but it seems that i'm not
> > doing that exactly because there have been produced about 200 trees
> > already.
> >
> > so what am I doing? :)
> >
> > all the best
> > pavlos
> >
> >
> >
> >
> >
> >
> > --
> > You received this message because you are subscribed to the Google
> > Groups "raxml" group.
> > To unsubscribe from this group and stop receiving emails from it,
> send

> > an email to raxml+un...@googlegroups.com <javascript:>.

> > For more options, visit https://groups.google.com/groups/opt_out

> <https://groups.google.com/groups/opt_out>.

> >
> >
>
> --
> Alexandros (Alexis) Stamatakis
>
> Research Group Leader, Heidelberg Institute for Theoretical Studies
> Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology
> Adjunct Professor, Dept. of Ecology and Evolutionary Biology,
> University
> of Arizona at Tucson
>

> www.exelixis-lab.org <http://www.exelixis-lab.org>

[David Kainer]

Thanks Tobias - that's what appears to be happening.