[Bioclusters] cluster hardware question

[bioinfo]

Kiran,
	I would like to suggest that you take a look at hardware
accelerators for your blast and HMM needs.  While a bit on the pricey
side, (academic pricing is a bit better) the performance gains for one
or two nodes is tremendous.  We have both a Linux cluster and a
TimeLogic Accelerator array and at this point no blast jobs are ever run
on the cluster.  At the moment we have only 2 accelerator boards, and
will probably add a few more in the future if funding permits.  Please
don't get me wrong, there are things that you can do with a cluster that
you cannot do with an accelerator but, in my opinion, for apps that are
going to be run over and over (Blast) you can't beat the performance
gain.  Not only is the management, space, and cooling requirements
tremendously reduced but the performance goes up.  You can probably
check around for specific stats about the number of white box nodes you
could replace with one accelerator card.  We decided to create a
heterogeneous computing environment to meet our various needs.  
	I am extremely happy with the accelerators we have purchased and
definitely recommend them.  If you would like some more info feel free
to contact me off list.

Mark 

-----Original Message-----
From: bioclusters-admin@bioinformatics.org
[mailto:bioclusters-admin@bioinformatics.org] On Behalf Of Kiran
Jonnalagadda
Sent: Thursday, January 16, 2003 9:22 PM
To: bioclusters@bioinformatics.org
Subject: Re: [Bioclusters] cluster hardware question


On Thu, 2003-01-16 at 17:35, Duzlevski, Ognen wrote:
> Kiran,
> 
> care to share your experiences? I am always keen on learning more so 
> that I don't make the wrong decisions :)
> 
> Are you a bioinformatics shop?

We're building clusters at Johns Hopkins University for Mascot, for
BLAST or a workalike, and for a Zope based protein database. I'm trying
to figure out if makes sense to have different clusters for each or a
single mega cluster.

The cluster is still just a plan so there are no experiences to share
yet. Will post here if we come across anything noteworthy.

Since none of us here are strong on the hardware side, what we're
picking off the list is invaluable knowledge. I hope we can add to it in
future.

-- 
Kiran Jonnalagadda
http://www.pobox.com/~jace

_______________________________________________
Bioclusters maillist  -  Bioclusters@bioinformatics.org
https://bioinformatics.org/mailman/listinfo/bioclusters