[R-sig-hpc] Rmpi on Mac OS X 10.5.8 dual core machine [typo fix + PS]

[Ross Boylan]

On Fri, 2009-12-11 at 16:01 -0800, Ross Boylan wrote:
> On Wed, 2009-11-18 at 17:09 -0200, Benilton Carvalho wrote:
> > Hi,
> > 
> > I decided to experiment with Rmpi on my dual core machine (Mac OS
> X  
> > 10.5.8).
> > 
> > So, I try the usual:
> > 
> > library(Rmpi)
> > mpi.spawn.Rslaves()
> > 
> > and let R "sit there"... I expected the R process to be idle, but
> for  
> > my surprise it uses, at least, 70% of CPU.
> > 
> > Is this an expected behavior? 
> I'd expect 100%, but that's based on a Linux cluster.
> OMPI does busy waiting, i.e., it runs in a tight loop checking for
> data.
> So it uses lots of CPU even when it's doing nothing.
> 
> The developers made this choice to increase the responsiveness on MPI,
> based partly on the assumed target use-case involving a cluster in
> which
> nodes were allocated entirely to the job running under MPI--i.e.,
> nothing else we need the CPU.
Should be "would need the CPU."

In a batch environment resources are generally reserved, so the
assumption is accurate.

We have a little cluster that doesn't fit that description; obviously
your laptop doesn't either.

Ross