CCL: ADF and SGE(Grid Engine) : How To ?



Dear Surasjajwej, 

This type of technical ADF question, is perhaps more suitable for the ADF forum: http://www.scm.com/forums/
or to send directly to our support E-mail address (support [at] scm.com). 

I will now forward this message to support, so you will receive instructions from one of my colleagues. 
Other than the generic recommendation to install the latest release ADF2009, I can only tell you now that the ADFjobs
command center in the ADF-GUI includes SGE support.

The GUI installation manual gives examples on how to set up a new SGE queue within ADFjobs: 
http://www.scm.com/Doc/Doc2009.01/Install/GUI/page1.html
http://www.scm.com/Doc/Doc2009.01/Install/GUI/page9.html

This allows ADF users to submit parallel ADF jobs to a remote cluster directly from the ADF-GUI on their desktop. 
This works in a transparent way and affords local visualization of the results. 

Best regards,
Stan van Gisbergen 

On Oct 30, 2009, at 10:03 PM, Sudarshan Wadkar wadkar^_^gmail.com wrote:

Dear CCL subscribers,
I have few queries regarding Grid Engine's queue system and ADF (2008.01e).
First a little background :
I have a 4 node mini-cluster(4 x Core 2 Quad) using Rocks 5.1/CentOS 5, where I have managed to install and run ADF2008.01e in parallel. The important file for parallel execution, ".machine" or the "machinefile" is working properly. It's contents are --

compute-0-0 4
compute-0-1 4
compute-0-2 4

I am confused about 3 things
1. ADF uses its own HP-MPI library ($ADFHOME/bin/hpmpi) how do I configure SGE so that it uses that HP-MPI library and not the system MPI ?
2. When I run ADF in interactive mode from ADF-GUI, I know I can change the .machinefile to control behaviour of ADF in a prallel run. How do I manage that using SGE? To put it in another way, how do I use SGE to determine which nodes are free and ADF should start as and when it gets, say 8 processors (i.e. 2 nodes) free? How to tell ADF to run only on those specific nodes?
3. How to create a parallel run shell script (which would interact with "qsub" command) from an ADf ".job" file?

Thank you :)


-Surasjajwej
~$udhi
"Success is getting what you want. Happiness is wanting what you get."
- Dale Carnegie
"There is only one success - to be able to spend your life in your own way."
- Christopher Morley

Dr. S.J.A. van Gisbergen
Chief Executive Officer
Scientific Computing & Modelling NV
Theoretical Chemistry, Vrije Universiteit
De Boelelaan 1083
1081 HV Amsterdam
The Netherlands                                
T: +31-20-5987626    
F: +31-20-5987629