[R-sig-Debian] rmpi/snow grabs all available CPU

[Faheem Mitha]

On Thu, 8 May 2008, Dirk Eddelbuettel wrote:

>
> On 8 May 2008 at 11:54, Faheem Mitha wrote:
> |
> | Hi,
> |
> | I'm testing affyPara on Debian lenny with
> |
> | R 2.7 from unstable, and the corresponding bioconductor packages from
> | bioconductor.org downloaded using biocLite as per usual.
> |
> | The command
> |
> | cl = makeMPIcluster(k)
> |
> | succeeds, but spawns k R slaves which promptly soak up all available CPU, which
> | is odd since they are not doing anything yet. This looks like a bug to me.
> | Similar behaviour is shown by a different machine, also running lenny,
> |
> | This is probably an issue with rmpi/snow. I notice that the Debian package   of
>
> I think what you see is actually Open MPI which polls all cpus all the time
> _on purpose_.  I am a bit hazy on the details but IIRC this may change in the
> upcoming 1.3 release.  CC'ing Manuel who may remember better than I do.
>
> | rmpi is now using openmpi. Any reason for not continuing to use lam as before?
>
> LAM is discontinued and the LAM authors recomemnd migration to Open MPI, see
> the LAM website.
>
> You can of course rebuild locally against LAM, or continue to use the old packages.
>
> | Should I submit a formal bug report for this?
>
> There is no bug here -- though I agree that I also dislike this Open MPI
> feature but it is in fact by design.  Check the Open MPI lists, it comes
> there a bit.
>
> Dirk

Hi Dirk,

Ok, but I still think that is pretty weird, not to say anti-social (taking 
up all the CPU for nothing, I mean). Plus the parallel call rmaPara takes 
much longer than the non-parallel version justRMA, which makes me wonder 
what is going on.

I think rebuilding against lam is not a bad idea for now.

                                                            Thanks, Faheem.