[Biodevelopers] Sun bits: solaris optimized blast binaries & a Gridengine-blast integration note on their website

[Chris Dagdigian]

Hi folks,

This forwarded email is from the Sun compbio SIG mailing list. Apologies 
in advance if someone already passed this on. I'm still brain dead from 
the ISMB/Brisbane meeting...

For people who don't care about the Solaris binaries the link to the 
HTML whitepaper on "Integrating BLAST with Grid Engine" is actually very 
good.

The paper gives a basic but step-by-step guide to setting up a NCBI 
blast aware Grid Engine installation that is not in any way specific to 
running on a Sun cluster or even using the commercial version of grid 
engine. If you are interested in trying Grid Engine this should be 
helpful -- particularly the shell scrips that are given showing how to 
submit blast jobs (with embedded SGE directives) into the cluster.

This could be something someone interested in getting an intro to Grid 
Engine could use to get things up and running on a Linux laptop or whatever.

The 'missing' info is how to actually configure, tweak and tune the Grid 
Engine configuration. The assumption in the whitepaper is that the user 
would just use the default queues and config provided by the SGE 
installation scripts.

In general this is OK -- the default queues are quite sensible but they 
generally come with 1 job slot per CPU which is sometimes good for blast 
and sometimes not. There are other parameters that one may want to tweak 
as well.

I'd recommend that people following this whitepaper look up the 'Howto' 
papers on http://gridengine.sunsource.net and read about how to create 
and configure grid engine 'complexes' and 'requestable/consumable 
resources'.

As an example of SGE complex usage for bioclusters -- When building 
similar systems with dual-CPU compute nodes I like to create a 'hog' or 
'greedy' complex that will allow users to request that  their job to 
have 100% exclusive acccess of a compute node.

With this setup a normal job would look like this:

$ qsub ./myBlastJobScript.sh

If I want my job to take exclusive use of a compute node with no other 
jobs running in the remaining job slot I can do this:

$ qsub -l greedy=1 ./myBlastJobScript.sh

Dumb example perhaps but complexes and resources are very very useful in 
SGE land and the whitepaper sorta skips over them.


-Chris




-- 
Chris Dagdigian, <dag at sonsorol.org>
BioTeam Inc. - Independent Bio-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

-------------- next part --------------
An embedded message was scrubbed...
From: Loralyn Mears <loralyn.mears at sun.com>
Subject: Solaris 32- & 64-bit optimized binaries for NCBI BLAST now posted!
Date: Thu, 10 Jul 2003 15:51:15 -0400
Size: 5590
Url: http://bioinformatics.org/pipermail/biodevelopers/attachments/20030712/70029821/Solaris32-64-bitoptimizedbinariesforNCBIBLASTnowposted.mht