[Bioclusters] Summary and Thanks for opinions on EST/cDNA assembly, clustering systems

Rayson Ho raysonlogin at yahoo.com
Mon Mar 7 12:33:50 EST 2005


I missed the whole discussion... too busy working on other stuff.
Something to add:

1) Make sure you benchmark YOUR code
Some CPUs show very nice benchmark scores, but they just fail to run
your code efficiently. I think most vendors offer at least remote
access to their testdrive machines.

2) Also, make sure you are comparing apples of the same class:
If you are planning to go for PBSPro, make sure you also try out the
commercial version of SGE. But nevertheless, ask on the Torque/PBS and
the SGE mailing lists before making up your mind.

The problem with clustering software is that your can't mix/match
PBS/SGE/LSF or Condor... If you go with PBSPro and later, you decide
that SGE is better, the $$ you spent on PBSPro will be lost :(

Rayson


--- Gary Van Domselaar <gary at www.bioinformatics.org> wrote: 
> I wanted to express my gratitude for all the valuable contributions
> to my
> query about how to best advise on the purchase of a computing system
> for
> EST/cDNA assembly and clustering. The question raised a lot of very
> interesting discussion and opinions, including some offline
> discussion.  
> That I can get such high quality advice from the top experts in the
> field
> sp quickly and for free, well it simply blows my mind--huzzah for the
> bioclusters list!
> 
> Here is how I interpreted the advice.  Hopefully I didnt misinterpret
> it 
> too badly.
> 
> 1. Don't put the cart before the horse.  Identify the expertise
> available
> to you and the workflow you intend to implement and use that to guide
> your
> hardware purchasing decisions.  This keeps you from looking like a
> grade
> 'A' dunce when the critical application you need to run is not
> compatible
> with the system on which you just spent 120 000 dollars. It also
> allows
> you to focus more on the science and less on the IT.
> 
> 2.  If you have a specific application, build your system around that
> 
> application.  Generic clustering systems may not meet your specific
> needs 
> and can cause lots of grief when you try to hack it to do your
> specific 
> task.
> 
> 3. Assembly and clustering require lots of memory to run efficiently,
> so 
> 64-bit SMP systems with lots of RAM are preferred.
> 
> 
> Anyways thanks again to all who contributed.
> 
> Regards,
> 
> g.
> 
> 
> 
> 
> 
> On Fri, 25 Feb 2005, Gary Van Domselaar wrote:
> 
> > 
> > Hey Gang,
> > 
> > I've been called in at the last moment to "consult" on the purchase
> of a
> > cluster for a sequencing project.  Admittedly, I know nothing about
> life
> > science clusters, despite having been subscribed to this list from
> its
> > inception.  I am not making any money on this consulting, just
> helping out
> > a neighbouring academic lab.  So what I know at this point is that
> they
> > have about $Cdn 200K to spend.  They hae already talked to Sun, and
> Sun is
> > offering them a "sweetheart" deal for (something) at about $120k. 
> My only
> > exposure is to a G4/G5 cluseter from BioTeam.  I am impressed with
> it
> > andit works really well for my purposes, and Im guessing it would
> work
> > well for theirs too.  I'm guessing a linux cluster would perfrom
> nicely
> > too. The lab currently does not have a bioinforatician, but I thnik
> they
> > have money for one.  I'll probably just end up pointing them to
> Glen, Joe
> > , and Chris, but any advice, suggestions, and pointers to where I
> can get 
> > a little more familiarity, well shucks, that would really be swell.
> > 
> > Decidedly,
> > 
> > g.
> > --
> > Gary Van Domselaar, PhD.
> > Postdoctoral Fellow, Computing Science
> > and Biological Sciences
> > University of Alberta
> > Edmonton, AB, Canada
> > Phone: 780-492-5969
> > 
> > 
> > Assistant Director, Bioinformatics.Org
> > gary at bioinformatics.org
> > 
> > 
> > _______________________________________________
> > Bioclusters maillist  -  Bioclusters at bioinformatics.org
> > https://bioinformatics.org/mailman/listinfo/bioclusters
> > 
> 
> _______________________________________________
> Bioclusters maillist  -  Bioclusters at bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters
> 


	
		
__________________________________ 
Celebrate Yahoo!'s 10th Birthday! 
Yahoo! Netrospective: 100 Moments of the Web 
http://birthday.yahoo.com/netrospective/


More information about the Bioclusters mailing list