[nycbug-talk] Memory sizing

Peter Wright pete at nomadlogic.org
Sat May 6 15:50:34 EDT 2006

> On May 6, 2006, at 12:52 PM, Isaac Levy wrote:
>> you with regard to code maturity and raw speed etc...  I think
>> without benchmarks, what we're really jawing about, is tradeoffs in
>> approach.
> It would be interesting to find out what google/amazon s3 are using
> -- both are advocating the use of clustered commodity hardware for
> speed/redundancy vs using solid, more expensive components that are
> faster on their own.

that's an interesting point.  i know google has a rather high expectation
that hardware will fail (and it does very frequently from what i've heard)
and their IT infrastructure takes this into accout.  in my current
environment we can not take that approach for various reasons (lack of
support staff being paramount).

i also think it really comes down a users specific use-case.  we have a
large group of people reading/writing huge data-set's 24/7.  for our
primary storage fiber channell SCSI backed NAS is needed for this.  our
secondary (mirroring/backup) storage is actually SATA.  it does not
provide the speed that SCSI has, but it is not needed in this case.  for
secondary storage we need lot's of cheap disk space.

i'll ping our filer people and see if they can offer any numbers on
failure rates of our SATA shelves versus our SCSI shelves....that might be
pretty interesting....


Peter Wright
pete at nomadlogic.org

More information about the talk mailing list