[Bioclusters] Research Study of Large, Computation-Intensive Grids

[Tim Cutts]

On 26 May 2005, at 7:09 pm, dave at princetonsolutionsgroup.com wrote:

> Thanks for your question!
>
> I know that there is a lot of confusion around how we are referring to
> these technologies.
>
> What I am looking for may be called a cluster in the sense that a 
> complete,
> 100+-node infrastructure of high-end, Xeon servers running Linux 
> resides at
> one location. It's probably connected to a SAN with fabric of Ethernet,
> Infiniband or Fibre Channel.

Low latency interconnects are currently fairly rare in genomics 
clusters.  The benefits don't justify the costs for a lot of our 
applications.  The same goes for SAN; SAN-attaching a cluster is very 
expensive indeed, for very little benefit.  Our 1000 node cluster uses 
ethernet for interconnects, and the SAN connectivity is only present on 
the fileservers.  As of now, we have no low latency interconnects such 
as Infiniband or Quadrics at all,

> And maybe they're running really sophisticated Celera or Accelrys
> applications?

Only a minority.  Most clusters that I've had anything to do with have 
run almost entirely open source or otherwise freely available software.

I last ran GCG (now owned by Accelrys), for example, back in 1998, and 
have used free alternatives ever since, and that's in both commercial 
and academic settings.

Tim

-- 
Dr Tim Cutts
Informatics Systems Group, Wellcome Trust Sanger Institute
GPG: 1024D/E3134233 FE3D 6C73 BBD6 726A A3F5  860B 3CDD 3F56 E313 4233