Skip to content

How to install R on a linux cluster.

10 messages · Dirk Eddelbuettel, Sean Davis, Hodgess, Erin +5 more

Dirk Eddelbuettel
#
On 21 April 2009 at 10:38, Richard Yanicky wrote:
| This might be common knowledge. I am looking to have a system admin install
| R on our linux cluster is there general documentation or a documentation
| resource that would help me direct him on how to do this? If additional
| information is needed please let me know.

On Debian / Ubunti all you need to get going is 

   $ sudo apt-get install r-base r-cran-mpi r-cran-snow

and you have R with MPI and the snow wrapper package. Not a bad place to
start as per the Schmidberger er al paper surveying parallel computing with
Linux.

There is no magic bullet -- Linux elements in a cluster are still 'just'
Linux machines.  So use the tools of the distro you (or your sysadmin) are
familiar with.  I like Debian/Ubuntu; you may want to look at Scientific
Linux which is Fedora-based but I am not sure how much R content you find
there. 

Dirk
Dirk Eddelbuettel
#
On 21 April 2009 at 10:25, Dirk Eddelbuettel wrote:
| and you have R with MPI and the snow wrapper package. Not a bad place to
| start as per the Schmidberger er al paper surveying parallel computing with
| Linux.

That was meant to say:  "... surveying parallel computing with R." -- sorry.
Benjamin Hofner
# 
There you go: http://epub.ub.uni-muenchen.de/8991/
(I think Dirk means this one)

Benjamin

Hodgess, Erin schrieb:
Damiano G. Preatoni
# 
In un messaggio del Tuesday 21 April 2009, Dirk Eddelbuettel ha scritto:
Great review. No mention of multicore, though.
I'm starting setting up an 8-core from Oracle (ops, I meant Sun), and I was 
wondering if there was really any advantage in multicore vs. snow/snowfall.

Any hints are welcome, will digest back to list.
Markus Schmidberger
# 
The multicore package  was released in January 2009, we did the review 
in end of 2008 )-:
Multicore has a great performance. I hope to present some results at the 
UseR2009 conference. At the moment I use the super-computer HLRBII at 
the LRZ in Munich, Germany (>9000 processors). There I have a 510 core 
multi-processor machine. Unfortunately we have some strange 
OS-load-balancing problems with the combination of R, PBS and Intel 
Compiler and more than 50 processors. Attached some first results with 
60 nodes. Up to 50 nodes snow and multicore have the identical 
performance. If you have problems which require a lot of memory and they 
can share the memory, than multicore has the better performance (snow 
than requires a lot of communication).

Best
Markus

Damiano G. Preatoni schrieb:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      A.J. Rossini
    

    

      
      #
    

  


  

      
On Wed, Apr 22, 2009 at 8:39 AM, Markus Schmidberger

        
<schmidb at ibe.med.uni-muenchen.de> wrote:

            

      
      
      
    

        
Only a small comment -- it would  be nice to have a multicore backend
(should be somewhat trivial) for SNOW....  (i.e. allowing the same
API, in the same way that you can use an NWS backend for SNOW  -- of
course, it would be nice to cleanup and modernize the SNOW api, as
well.

<ducking type=quickly, since I'm not going to do it right now/>

best,
-tony

blindglobe at gmail.com
Muttenz, Switzerland.
"Commit early,commit often, and commit in a repository from which we
can easily roll-back your mistakes" (AJR, 4Jan05).

Drink Coffee:  Do stupid things faster with more energy!