On Feb 22, 2012, at 11:52 PM, Cary Gordon wrote:

> EC2 works for a lot of models, but one that it does not work for is
> small traffic apps that need to be available 24/7. If you have a small
> instance (AWS term) running full time with a fixed IP, it costs about
> $75 a month. If you turn it on for 2 hours a day, it costs about
> $15/month. A large instance is about $325.
> Now where it gets interesting is if your app needs a large instance,
> but only run a few hours a month, you might be able to run a micro
> instance that is set to start a large (or ???) instance on demand, and
> run the whole thing for peanuts.

We've looked at something similar (not Amazon, NASA is working on its
own cloud service) where we'd locally run a server, but at times of
high demand, pass off to the cloud service.

If you have applications that are cyclic, I could see it being an
advantage to have something take over in the peak times.  Eg,
when I worked for a university, the system we used for class
registration was okay ... not great, but okay ... but the incoming
freshmen were brought in in 3 or 4 'orientation' periods over the
summer, and they'd all hit the system on the same day, at the same
hour (well, 1/3? 1/4 of the incoming class)

The system performance went to complete crap.  We're talking about
throughputs worse than if we had metered the access.  (and the
DBAs refused to look at database tuning, insisting that it was a
webserver problem ... it was of course, a database issue, but it
was months before we got it straightened out)

I could see conferences using something like this -- where almost
all of their traffic is on the days of deadlines, or during the
conference itself.

If the load's pretty uniform, I don't think their pricing model
is all that advantageous.  (and I have no idea how they handle
the loads over christmas, as the reason for the cloud is to make
money back on their excess capacity they need for the christmas
sales period.)