Great thread!

At WFU we used reserved AWS instances which lowered our overall costs
but committed us to the amazon platform for a year.  We also wound up
grouping most of our services on a large server (~$87 per month after
reservation fee) so that we could take advantage of all of that
capacity.

Our infrastructure included 3 servers and about 500 GB of storage
(Large production server with 90% of library services, 1 small server
for High density storage system, 1 small server for
puppet/monitoring/documentation).  The reservation fees for these
servers were around $1380 per year and we paid approximately $275
total per month for computing costs and disk space. Data transfer and
other costs were minimal and are included in the $275.  A rough yearly
cost for these services comes to $4680.  A bit more than we were
looking at for physical server costs (3 servers for $4000 each with 3
years paid support) but these costs meant that we had a lot more
flexibility than we would have had with three servers sitting in our
campus IT datacenter (without root access).  As a side note, we found
that storage space was more expensive than CPU time and wound up
keeping our multi-TB storage array on site instead of in the cloud.

If I was re-building this today I would explore some other options -
RackSpace (Cheaper CPU time), RightScale (Automated server
configuration/deployment), Heroku/Google Apps Engine (free PaaS
levels) and focus on getting at least a more robust infrastructure at
the same cost (if not with some savings).  Rackspace for example
offers their smallest server at $.015 per hour without reservation
fees and RightScale offers a free support level that could work well
for small/medium sized libraries.

FWIW, when I was pulling together numbers for this email I noticed
that Amazon has changed their reservation fees and pricing model.
Depending on which reservation fee I selected I either saved about
$600 per year or spent $300 more per year using the same
infrastructure discussed above (http://aws.amazon.com/ec2/pricing/).
If you are really interested in cost calculations and ROI, some
helpful resources include Yan Han's recent work in ITAL comparing
real-world cloud computing costs -
http://ejournals.bc.edu/ojs/index.php/ital/article/view/1871/1709 and
Chapter 3 of George Reese's book "Cloud Computing Architectures" in
which he explores some approaches to calculating ROI for cloud
services.

Erik

Erik Mitchell
Assistant Professor
College of Information Studies
University of Maryland, College Park
http://erikmitchell.info, http://ischool.umd.edu

On Thu, Feb 23, 2012 at 12:38 AM, Tim Spalding <[log in to unmask]> wrote:
> We did some tests on it, but found it a very poor fit for a site
> dependent on huge amount of data which much be "present" to the
> basically the whole system all the time and up-to-date. In other
> words, we found it didn't match a site based on MySQL slaves
> replicating here and there, and with memcached needing to be spot-on.
> Under some circumstances we'd consider shuffling some image rendering
> and delivery tasks to it, but that's about it.
>
> Tim
> LibraryThing