[Bioclusters] LSF 4.2 w/ Redhat 7.2 (2.4.X kernel with glibc2.2) install problem

Chris Dagdigian bioclusters@bioinformatics.org
Tue, 30 Apr 2002 19:57:21 -0400


Hey Ron-

I like SGE and play with it quite a bit on my laptop and elsewhere. It 
is my DRM of choice for people who cannot afford LSF or have basic basic 
workflow needs where there is no need for LSF's fancy features.

However, Platform LSF is still superior to GridEngine and I recommend it 
in places where it makes sense. For this particular project (Bauer 
Center for Genomics Research @ Harvard University) Platform LSF was the 
best choice given their needs.

I recommend what I think is best for the customer and don't let 
philosophy or my reputation as an open source zealot get in the way :)

If anybody wonders this is why I recommended Platform LSF for this project

o LSF has the lowest administrative burden period. This was the #1 
consideration given that the CGR will initially have a very tiny IT 
staff. I can set things up over there and walk away knowing that anyone 
with basic linux knowledge can keep the system up and running. The cost 
of the LSF license is far lower then what it would cost to hire an 
additional admin (or keep me onsite at an hourly rate :) heh.

o LSF is the most stable DRM and has the best redundancy and failover 
features out of any DRM system I've been involved with. GridEngine is 
getting better at this kind of stuff and I really like what I've been 
reading about the proposed new failover stuff with SGE shadow_master but 
its not there yet.

o Customer has a need for eventual parallel environment support. I 
really like the "tight" PE environment that LSF Parallel provides 
instead of the half-assed "we'll generate a custom mpihosts file on the 
fly for you but nothing more" approach that the "loose" PE systems do. 
I've learned from bitter experience over at Vertex Pharma that the DRM 
needs to be able to track, log and control the various child processes 
spawned off by parallel MPI jobs. Otherwise things can spin wildly out 
of control. I think that SGE has good tight integration with PE 
environments so this one may be a wash...

o The ensembl genome annotation pipeline over at the Sanger Centre was 
written with LSF as the native DRM. Harvard eventually wants the ability 
to take full advantage of what those crazy ensembl.org ninjas have created.

o Possible nifty 'data grid' stuff happening w/ Avaki. I'd eventually 
like to make Harvard CGR a node on some sort of Avaki-enabled data grid. 
This may also be a wash as SGE will likely add grid functionality just 
as fast as everyone else. There seems to be lots of cool stuff happening 
over in the SGE camp with  real grid stuff.

-Chris




Ron Chen wrote:

> Are you handling a customer's cluster?
> 
> I remember you said that SGE is as good as LSF,
> however, SGE is free, and opensource.
> 
> Wouldn't it be nicer to install SGE?
> 
> -Ron
> 



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