[Beowulf] Re: Beowulf Digest, Vol 30, Issue 7

[David Mathog]

Mike Davis wrote:
> 
> I don't mean  to hijack the thread, but if Dave's users can fit the db's 
> that they are running (Blast for instance against) in /tmp on the 
> compute nodes, overall performance increases.

That's more or less how it works, except the databases are sliced up
across the nodes and the results stitched back together on
the one machine where the jobs are submitted (which for historical
reasons is an older Solaris machine, not the file server). 

The file server wears a fair number of hats - none red though ;-):

It is the head node for the compute cluster.  (runs DHCP, SGE, NIS,
  etc. servers,  holds images of nodes, and so forth.)
It is the file server for a bunch of interactive users on other
  machines (NFS and Samba mounts).
It accumulates and distributes results from the DNA sequencing
  facility.
It downloads databases from the NCBI (and elsewhere) and keeps one
  "intact" copy on hand.  However that isn't used directly, the
  file server processes that into slices and pushes it out to the
  compute nodes.  That processing tends to be IO bound (the logic
  isn't complex and there's no floating point involved at all.)
It will also do ??? in the future (ie, we tend to pick
  up new work without a corresponding budget to buy new hardware.)


Regards,

David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech