[R] Some hpc problems with doMPI and runmpi (I know there is R-SIG-HPC)

[Pascal Oettli]

Hello,

In your email, you speak about foreach()%dopar%, but in your script,
it is foreach()%do%.

Best,
Pascal

On Thu, Aug 7, 2014 at 7:19 PM, André Ziervogel
<Andre.Ziervogel at psychol.uni-giessen.de> wrote:
> Dear R people,
>
> I’ve been doing some hpc using R and openmpi. Unfortunately I’ve encoutred a major problem and it’s nature is hard to pin down:
>
> Essentially I call mpirun Rscipt … as soon as the script reaches a foreach()%dopar% it halts indefinitely. I’ve attached the qsub script:
>
> #!/bin/bash
> #$ -S /bin/bash
> #$ -N test_14
> #$ -cwd
>
> #$ -V
> #$ -o "/fhgfs/g61570/Spectral Databases/log/test_14_"$JOB_ID
> #$ -j y
> #$ -q regular
> #$ -pe openmpi 8
> #$ -l h_rt=00:15:00
> #$ -l h_vmem=1.9G
> #$ -m eas
> #$ -M andre.ziervogel at psychol.uni-giessen.de
>
> module add gcc
> module add openmpi/gcc/64/1.6.5
> module add R/gcc/3.0.1
>
> date                       #log start time
>
> echo "Number of slots " . $NSLOTS
>
> mpirun Rscript /fhgfs/g61570/Spectral\ Databases/test_10.r > /fhgfs/g61570/Spectral\ Databases/log/test_14.Rout
>
> date
>
> exit
>
> and the R file:
>
> suppressMessages(library('doMPI'))
>
> skylla.cluster <- startMPIcluster()
> registerDoMPI(skylla.cluster)
>
> cat(paste("COMM SIZE: ", mpi.comm.size(0), " cluster size: ", clusterSize(skylla.cluster), "\n",sep = ""))
>
> tmp.time <- proc.time()
> sample <- foreach(i=seq(from=0, to=1000, by =1),.combine='c',.inorder=TRUE) %do%
> {
>         r <- sqrt(i^2 + i^2) + .Machine$double.eps  * factorial(i)
>         sin(r) / r
> }
> cat(paste("Processing seriell time: ", "\n", sep = "  "))
> print(proc.time() - tmp.time)
> #print(sample)
>
> tmp.time <- proc.time()
> sample <- foreach(i=seq(from=0, to=1000, by =1),.combine='c',.inorder=TRUE) %dopar%
> {
>         r <- sqrt(i^2 + i^2) + .Machine$double.eps  * factorial(i)
>         sin(r) / r
> }
> cat(paste("Processing parallel time: ", "\n", sep = "  "))
> print(proc.time() - tmp.time)
> #print(sample)
>
> closeCluster(skylla.cluster)
> # mpi.close.Rslaves()
> # mpi.exit()
> mpi.quit(save='no‘)
>
> Any suggestions would be highly appreciated! Thanks!
>
> Best
>
> André
>
> ------------------------------------------------------
> Dipl. Psych André Ziervogel
> andre.ziervogel at psychol.uni-giessen.de
> ------------------------------------------------------
>
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>



-- 
Pascal Oettli
Project Scientist
JAMSTEC
Yokohama, Japan