[BioBrew Users] 'R Project' using parallelism

[adolfo rafael martinez]

Dear BioBrew Users;

I would like to share the solution I found for 'R project'. It was
implemented at the UT Center for Computational Biology and
Bioinformations. But, there is still a need to automate the PVM daemon
startups on the compute nodes. So, any feed back would be appreciated.

I found James Burnash at Lucent Technologies (NJ) was also working on
this. After combining efforts, a solution was found.

This was done on a cluster primarily made of Dell PowerEdges 1850's
(and a 2850 head node).

Using x86 binaries, thinking it's a safe initial choice, turned out to
be a mistake. So, the following i686 binaries were used:

R-2.3.0-2.rh4AS.i686.rpm
libRmath-2.3.0-2.rh4AS.i686.rpm

The rpms were added as described in

"5.1. Adding Packages to Compute Nodes"

But, I had to manually start the pvm daemon on each node.

I tested "R Project" with

library(snow)
cl <- makeCluster(2, type = "PVM")

I didn't have success with type MPI. So, type PVM was used.

And, the following example was used to prove functionality

http://www.sfu.ca/~sblay/R/snow.html

Now, I just need to automate the startup of the PVM daemons on each
node, and make certain jobs can use the SGE qsub mechanism.

If anyone has any tips, they would certainly be appreciated.