Re: mpi problems on amd64
My original question was based on the mistaken assumption that I was on
a node running Lenny. In fact it was etch, and the lenny version of
rmpi use openmpi (NOT lam, as I said earlier).
However, I'm still having troubles with Rmpi on lenny (amd64); I'm
getting
[n7:04654] mca: base: component_find: unable to open osc pt2pt: file not
found (ignored)
when I run under R without invoking mpirun. When I use mpirun with R I
get nothing at all.
The message about pt2pt is odd, since the changelog shows that problem
was fixed. Maybe the fact that it's not in a proper MPI environment is
contributing.
I suspect something about my local setup, or the fact that the saved
image was run under lam, is behind this.
Ross
On Wed, 2009-04-15 at 13:55 -0700, Ross Boylan wrote:
> First, thank you for getting rmpi working with openmpi.
>
> Second, I have a question. The r-cran-rmpi changelog (0.5-6-3) mentions
> a problem with openmpi and amd64. Do you have any sense of whether this
> would be an issue for the Lenny version of lam? I know openmpi is a
> descendant of lam.
>
> Also, is the problem likely to have effects for uses other then rmpi?
>
> We have a new cluster that is migrating toward Lenny (currently most
> nodes are etch) and is amd64 architecture (Xeon, really). Since the
> lenny version of rmpi uses lam, a lam/amd64 problem might bite us.
>
> It looks as if using rmpi with openmpi would require a fairly
> significant chunk of unstable (certainly R and openmpi, maybe more
> libraries they depend on). I think our sysadmin would prefer to keep
> things more stable.
>
> Thanks.
> Ross
>
>
Reply to: