[sf-perl] episodic cloud-based computation?

[James Briggs]

Hi Rich.

Cloud linux instances are usually billed by the hour, which doesn't match
your computing granularity of two seconds.

Here's some more details ...

Specifically:

- "Partial instance hours consumed are billed as full hours" on AWS
  See http://aws.amazon.com/ec2/#pricing

- typically it takes minutes to startup and terminate instances on
  AWS, overwhelming your actual usage of two seconds.

- each vendor uses their own API for creation and termination of
  instances, so the portability that you imply doesn't exist
  unless you use RightScale ($500/month) or something else

- If you keep your instance around and running, then it would be billed
  for the entire duration of that.

Some alternatives that involve billing at the granularity you want are:

- The closest thing to what you want is prolly Google AppEngine, but
they don't run Perl programs.
- You can get a hosting account on a large host like GoDaddy, and run your
jobs as CGI programs across their load-balanced web clusters.
- Some people have also exploited other SaaS APIs, like Salesforce.com,
as low-cost compute farms.

Thanks, James.

On Sun, 16 Oct 2011 10:54:48 -0700, Rich Morin wrote
> I have a prospective need for some episodic cloud-based
> computation.  For example, I might need one or two dozen
> Linux instances for about two seconds each, but not need
> them (at least initially) more than a few times per day.
> 
> My plan is to use Chef to set things up, so I should be
> pretty flexible about vendors, etc.  Given all of this,
> what cloud providers are likely to be able to give me a
> good deal for keeping my instances around and running
> them at need?
> 
> -r