[Bioclusters] Xserve G5 memory

[Chris Dwan]

> We have 2 gigs of RAM per Node.  For our testing purposes we are 
> willing to
> add as much RAM as necessary, up to 8 is permitted on the Xserves.  
> The IBM
> actually has 8 gigs in it.  I'd be willing to tweak the kernel, or 
> anything
> else necessary to have memory allocation work a little more 
> effectively for
> BLAST if possible.  We've even considered simply dedicating a machine 
> or two
> to performing at its best with BLAST.

It might be wise to look into some of the specialized accelerators 
available on the market.  If BLAST is really the primary and only use 
for a certain set of resources, it's possible to achieve two or three 
orders of magnitude of speedup by using specialized hardware.

Of course, it all depends on how the problem is stated and what the 
users need.  I know very few researchers for whom BLAST is the rate 
limiting step.  It's a mature, popular, free piece of software, which 
makes for a great benchmark, but my suspicion is that even if the time 
requirement for BLAST were reduced to zero, it wouldn't appreciably 
change the way in which life scientists do their work.

Naturally, many of the exceptions are on this list.  :)

-Chris Dwan
  The BioTeam