[Rd] Rmpi on Linux (SuSE) with OpenMPI

[Dirk Eddelbuettel]

(not a question for r-devel; you are simply having (local) troubles with a
package) 

On 5 September 2008 at 10:05, giuseppe1.milicia at hsbcib.com wrote:
| 
| Guys,
| 
| I'm having troubles with Rmpi on Linux. I got it to work on windows, but on
| Linux (SuSE 64 bit), using OpenMPI, I can't seem to get it to work :(
| 
| When I do library(Rmpi) I get the warning:
| 
| mca: base: component_find: unable to open osc pt2pt: file not found
| (ignored)

It's an Open MPI warning you can suppress in the Open MPI configuration.

| Interestingly enough all the examples from the snow tutorial (
| http://www.sfu.ca/~sblay/R/snow.html) work despite the warning.

Yes because it is just a warning about additional functionality you are not
using.

| But then, when I run my app I get
| 
| mca: base: component_find: unable to open osc pt2pt: file not found
| (ignored)
| mca: base: component_find: unable to open osc pt2pt: file not found
| (ignored)
|         2 slaves are spawned successfully. 0 failed.
| *** Process received signal ***
| Signal: Segmentation fault (11)
| Signal code: Address not mapped (1)
| Failing at address: 0x1c
| *** Process received signal ***
| Signal: Segmentation fault (11)
| Signal code: Address not mapped (1)
| Failing at address: 0x1c
| [ 0] /lib64/tls/libc.so.6 [0x2a95c87500]
| [ 1] [0x2a99b6df4c]
| *** End of error message ***
| [ 0] /lib64/tls/libc.so.6 [0x2a95c87500]
| [ 1] [0x2a99b6e28c]
| *** End of error message ***
| 
| Google has brought no useful info... The only discussion on the subject is
| about an older version of Rmpi (I checked, the fix is in my version). For
| reference that discussion is here
| http://www.open-mpi.org/community/lists/users/2007/10/4185.php
| 
| I'm running on a multiple cpu box rather than a real cluster, that should
| make things easier though... Perhaps some permission issue?

I don't think so. You are getting segfaults. That indicates a code bug or
maybe mismatch between libraries.

But as you didn't provide a reproducible example so we have no way of help
you.

Now, as you say 'multiple cpu box rather than a real cluster' why don't you
stick to using snow with sockets bypassing all your MPI issues ?

Hth,  Dirk

-- 
Three out of two people have difficulties with fractions.