[R] snow apparently using only one CPU (on Mac OS X)

Santiago Beguería Portugués santiago.begueria at csic.es
Thu Jul 26 17:27:24 CEST 2012


Hi,

I am trying to optimize a complex model calibration using a genetic algorithm. I choose rgenoud package because it allows for easy parallelization through snow package. I am on a Mac Pro with 2 x 2.66 GHz 6-Core Intel Xeon machine, i.e. I have 12 CPUs available. 

So I set the cluster options as:

setDefaultClusterOptions(rshcmd='echo ', user='> /dev/null', master='localhost')

then I start a SOCK type cluster with:

cl <- makeCluster(rep(';',12), type='SOCK')

and then I use:

gnoud(..., cluster=cl)

Then when I check the activity of the CPUs I realize that 12 processes have already started, but the CPU load of each process is extremely low (<4% most of the time). If I use other cluster configurations (i.e., a different number of CPUs) the CPU load per process increases inversely proportional to the number of CPUs. I believe that snow is only using one CPU and sharing the computing time within all the processes. Does anyone have the same experience with snow?

Cheers,

	Santiago Beguería



More information about the R-help mailing list