[GE users] Documentation about SGE

Chris Dagdigian dag at sonsorol.org
Fri Jul 23 16:10:46 BST 2004


{my $.02 of course, your experience may vary}

I agree with Vladimir's comments about OpenPBS. If you make a decision 
to use PBS you should use PBSPro irregardless of your status as a 
commercial or academic institution. The differences are significant and 
PBSPro is superior to OpenPBS.

I'll disagree gently with Vladimir's comments about LSF :) In my 
experience using LSF and SGE heavily over the last few years I've found 
that overall LSF edges out SGE as the 'better' product if you rule out 
the cost. heh.

LSF has the following advantages:

o lower administrative burden (subjective, I know). In general I find it 
takes "less work" to manage and debug a LSF cluster, particulary if you 
need to implement sophisticated resource allocation policies. This can 
be a significant thing -- the cost of purchasing LSF in some cases can 
be justified entirely by the admin/staff time savings.

o LSF *by far* has fault tolerance and resiliancy features that blow 
away all the competition. The SGE shadow master does not come close to 
LSF's ability to keep "electing" a new master node as systems fail or 
drop offline one by one.

o For the corporate types with big budgets the various add-on modules 
for LSF that enable WAN-scale job distribution and system accounting are 
really really good.  There are many powerfull add-on products.

There are more reasons but this is not the right forum.

Lest people think I'm spreading anti-SGE FUD I should point out that I 
use, evangelize and recommend SGE daily and that currently in my 
consulting practice, SGE is the software that I deploy "first". LSF is 
the 2nd choice and is used when required.


Ok back to SGE vs PBS which was the real question asked here

I don't use PBS. Ever. The dealbreaker for me when I last looked at PBS 
was the fact that it did not support Array Jobs which are a very 
critical feature in the industry where I work. My last look at PBS was 
several years ago so I could be obsolete with this opinion.

If you need to run an application thousands and thousands of times with 
only minor changs in the command-line arguments and input files then 
Array Jobs are something you need to have.

The other thing was that (subjective again) it felt that managing the 
PBS cluster and keeping it from falling over or losing jobs took "more 
work" than either SGE or LSF. Administrative overhead is a big deal for me.

The final scary point was that I have not yet met a serious, production 
user of OpenPBS who has not patched or otherwise modified the PBS 
sourcecode to fix things.


Regards,
Chris


-- 
Chris Dagdigian, <dag at sonsorol.org>
Independent life science IT & informatics consulting
Office: 617-666-6454, Mobile: 617-877-5498, Fax: 425-699-0193
PGP KeyID: 83D4310E Yahoo IM: craffi Web: http://bioteam.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list