Using R in a linux cluster

2 messages · Paco Pastor, Dirk Eddelbuettel

Tue, Jun 15, 2010 7:02 AM #

Hello

I'd like to use R in a Ubuntu cluster so I have installed R and Rmpi in 
a clean Ubuntu Lucid installation. Reading about Rmpi I need to have R 
and Rmpi installed in all nodes in the cluster. I have tried to export R 
executable and /usr/lib/R to the other nodes. When I try to run R in 
node2 I get the following error

meteo at boira2:/boira/R$ ./R

./R: line 208: /usr/lib/R/etc/ldpaths: No such file or directory

/usr/lib/R/bin/exec/R: error while loading shared libraries: libR.so: cannot open shared object file: No such file or directory



Then, after exporting /etc/R I get a the same error with libblas.so.3gf 
not found

What should be better? to install R and Rmpi in every node or to export 
R and all dependencies from the master node to the rest of nodes in the 
cluster? I'd prefer to keep R installed just in the master node so any 
packages update or config changes are not to be repeated in the other 
nodes (just 2, not a big deal but...)

Thanks in advance.

---------------------------------------------------------------
Francisco Pastor
Meteorology department
Fundaci?n CEAM
paco at ceam.es
http://www.ceam.es/ceamet - http://www.ceam.es
Parque Tecnologico, C/ Charles R. Darwin, 14
46980 PATERNA (Valencia), Spain
Tlf. 96 131 82 27 - Fax. 96 131 81 90
---------------------------------------------------------------
Usuario Linux registrado: 363952

Dirk Eddelbuettel

Tue, Jun 15, 2010 7:37 AM #

Hi Paco,

On 15 June 2010 at 16:02, Paco Pastor wrote:

| I'd like to use R in a Ubuntu cluster so I have installed R and Rmpi in 
| a clean Ubuntu Lucid installation. Reading about Rmpi I need to have R 
| and Rmpi installed in all nodes in the cluster. I have tried to export R 
| executable and /usr/lib/R to the other nodes. When I try to run R in 

No, that won't work.

[...]
| Then, after exporting /etc/R I get a the same error with libblas.so.3gf 
| not found
| 
| What should be better? to install R and Rmpi in every node or to export 

Yes. 

All nodes need to 'symmetric' ie have R, the same packages, libraries, ...
Even the paths should be identical.

| R and all dependencies from the master node to the rest of nodes in the 
| cluster? I'd prefer to keep R installed just in the master node so any 
| packages update or config changes are not to be repeated in the other 
| nodes (just 2, not a big deal but...)

A common solution is to have the master either be an NFS server or write to
an NFS server.  Eg I have one /usr/local/lib/R/site-packages/ and the other
machines access this via NFS.  Pretty easy when you use autofs auto-mounting.

But as so often, you need an extra bit of sys.admining to get all this going.

Regards, Dirk