[Beowulf] commercial clusters

[Buccaneer for Hire.]

--- Stuart Midgley <sdm900 at gmail.com> wrote:

> I've spoken quite extensively to several
> organisations here in  
> Western Australia who each have 1000+cpu systems in
> Perth, 2000+cpu  
> systems in London and 4000+ systems in Houston (oil
> and gas seismic  
> processing).  They tell me that they have tried to
> sell just the  
> cycle time to no success.  The value is in the
> specialised services  
> they offer (in these cases seismic processing),
> which just happen to  
> use a lot of cpu's.
> 
> These companies even go as far to roll their own
> cluster file systems  
> (using FUSE+nfs) and writing their own
> queuing/scheduling environment  
> etc.  The higher level services certainly pay well
> :)
> 
Hi Stu and friends,

Seismic is a totally different beastie.  He probably
does not want to do that.

You have very serious performance parameters you need
to maintain.  The sheer size of data I/O is mind
numbing for those not working in the field.  By the
time you start with 5+TB of data, travel time data
(needed on even higher performance storage), space for
temporary files and results, the 400-1000 nodes that
will be running hard over the next 4-6 weeks...  And I
just thought the procurement costs were hard.  Add to
that infrastructure to support and AC, security, etc. 
You have spent a couple of fortunes.


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com