[Bioclusters] SMP system for bioinformatics apps

[Chris Dagdigian]

2 years ago it was pretty easy to make the claim that Opterons had the  
price/performance advantage for the general landscape of life science  
related codes.

The new Intel products have narrowed or removed the gap and now we are  
in the realm of "you must benchmark your own most critical apps" if  
you want the answer that best fits your needs. There is no easy once  
size fits all answer any more. Sounds like you know exactly what apps  
are important for you so that makes things easier if you want to try  
some pre-purchase benchmarking.

I'd discourage you from building your own systems, the price delta  
between DIY and systems from vendors is small enough now that it's  
often better to pay a few percent more and get a system that comes  
with some engineering and warranty support behind it.

For vendors you should first consider what relationships your school  
has. It would be quite common in a academic setting for there to be  
some deal with Dell, IBM or whatever such that you could get hardware  
from this vendor and gain additional warranty or service entitlements  
"for free" as part of an institutional agreement. Pricing might also  
be better if you go through the campus sales rep.

If there are no such agreements or the chosen "big" vendor does not  
offer the CPU variety you may want to look at then I'd recommend  
looking at Silicon Mechanics (http://www.siliconmechanics.com/) - I've  
used them for single server and small-cluster orders and they have a  
good mix of Intel and AMD product offerings. One of the things that  
drew me to them in the first place is their web configuration system  
which for all but the newest products can give you nice real-time data  
about electrical consumption and heat output for the product you are  
putting together. It's nice to know the thermal/power data and  
sometimes this is hard to get out of other vendors.

Just my $.02

-Chris


On Sep 19, 2008, at 12:39 PM, Chuck Pepe-Ranney wrote:

> Hello Biocluster Community,
>
> My lab is interested in purchasing a SMP system to run various
> bioinformatics applications.  At the moment we heavily use Arb,
> NCBI-BLAST/WU-BLAST, and RAxML.  We also sparingly use ClustalW and
> MrBayes.  We will have access to a cluster at our institution in the  
> near
> future so we're not interested in purchasing a traditional cluster  
> of our
> own.  We're hoping that a multi-socket multi-core system in its  
> simplicity
> will be easier to maintain and administer than a cluster.  Obviously  
> we'd
> like to make sure that our hardware meshes as best as possible with  
> the
> structure of the programs we plan on using.  RAxML and Arb would be  
> our
> highest priorities at the moment.  BLASTing against large databases
> hopefully will be done on the institution's cluster using mpiBLAST.   
> We will
> be running NCBI-BLAST or WU-BLAST jobs on our system with smaller  
> databases,
> however.  We're hoping to to gain some insight by studying some  
> existing
> systems and are wondering if any members of the bioclusters group  
> had some
> relevant experience or expertise to share?
>
> At the moment we're invetigating CPUs.  A collaborator is heavily  
> endorsing
> the Intel Xeons but RAxML scales better on AMD Opterons.  Has Intel  
> come up
> with any recent innovation to compete with the Opteron memory  
> organization?
> Also, any help with subsequent steps involved would be appreciated.   
> We're
> considering building this system ourselves but would welcome advice  
> as to a
> particular vendor to contact if we decide to go with a pre-built  
> solution.
>
> Thanks,
>
> - Chuck
>
> -- 
> Chuck Pepe-Ranney
> Graduate Student
> Environmental Science and Engineering Division
> Colorado School of Mines
> Golden, CO 80401
> Email: cpepera at mines.edu
> office: (303)384-2218
> cell: (575)313-0993
> _______________________________________________
> Bioclusters maillist  -  Bioclusters at bioinformatics.org
> http://www.bioinformatics.org/mailman/listinfo/bioclusters