[R-sig-hpc] Rmpi working with OpenMPI and PBSPro but snow fails

Huw Lynes lynesh at cardiff.ac.uk
Thu Mar 5 12:24:44 CET 2009


OK results of some more testing. RMPISNOW is looking for
OMPI_MCA_ns_nds_vpid which OpenMPI 1.3 doesn't seem to export. It does
export OMPI_MCA_orte_ess_vpid. So I've altered RMPISNOW to look for
that.

This results in one R process being launched as --no-save and the others
being launched as --slave.

However the slave processes all exit leaving the master hanging there
with nothing to do.

To debug I replaced --slave with --verbose --no-save
The last thing the worker processes print is 

 >R_ReplConsole(): before "for(;;)" {main.c}

At this point it seems like it would be worth going back to basics and
trying to get snow working with the SOCK method rather than mpi. 

So using this simple script:
library(snow)

ncpus <- Sys.getenv("NCPUS")
cl <- makeCluster(ncpus, type = "SOCK")
clusterCall(cl, function() Sys.info()[c("nodename","machine")])
stopCluster(cl)


fails with:
Error in makeMPIcluster(spec, ...) : no nodes available.
Calls: makeCluster -> switch -> makeMPIcluster
Error in inherits(cl, "cluster") : object "cl" not found
Calls: clusterCall -> checkCluster -> inherits
Error in stopCluster(cl) : object "cl" not found

It seems to use MPI if it can find it even when I've explicitly told it
to use sockets. Is there a way around this?

Thanks,
Huw 



-- 
Huw Lynes                       | Advanced Research Computing
HEC Sysadmin                    | Cardiff University
                                | Redwood Building, 
Tel: +44 (0) 29208 70626        | King Edward VII Avenue, CF10 3NB



More information about the R-sig-hpc mailing list