Computing as a utility is a very powerful analogy. Jeff Bezos
has talked about (in
a recent article in wired and also a presentation @
Startup School) in regards to their Amazon Web Services (AWS).
Data centers are table stakes. You need a really good data
center to provide software and other services. But this is table
stakes: you need it but you can't differentiate on it (at least,
not much to the end consumer).
Amazon was scaling up and was thinking about how they were going
to do it. They had a clever idea and then realized that "if it was
helpful to us, it would be helpful to others." Thus was born,
AWS.
Competition Notes
"Out-googling Google" some people have said. And Microsoft (and
Google and Rackspace) should be scared
(and I think they are - not
that I have any inside info). Most of these companies have
competing offerings. Microsoft just launched their platform: Live Mesh (don't be confused, Live
Mesh is a platform, but their current "experience" [aka
"application"] is a file syncing & sharing offering). Rackspace
offers server - real, live, so-2007-physical, servers that you can
touch. Google has their "App Engine" that is similar to Amazon's
EC2.
Microsoft in particular should be awake because it's so hard to
run their software on the EC2. I've been looking at running a
Windows Server on the Amazon EC2, but there are few options and
they aren't great. I'm sure there are good reasons for this, but
not everyone can't switch off Microsoft and not having a compelling
option to do so doesn't help sales.
Where this might go
I don't for a minute think that no one will have their own data
centers. There certainly will be. But they may look more like
today's self-generated electricity: backups. Or maybe, you'll end
up with several utility computing companies where you can buy your
services and you'll differentiate your "source" in that way.
But thinking about computer services like utility, other ideas
become apparent.
1. Variable Pricing
There is a push with electricity companies to charge different
prices for electricity at different times: based on demand1. This
often involves them sending information to homes via color-coded
lights indicating current demand, so users can respond
accordingly.
Imagine this for the elastic compute cloud. The compute cloud is
already measuring all usage. I'm sure their systems already monitor
usage in the aggregate. Programming a system to allow for scaling
up and down based on current pricing would make a lot of sense.
Some companies want the computing when they need it but some don't.
Consumer-face web applications need it at specific times (when a
customer visits) while Scientific endeavors (e.g. Blue Origin) don't.
So if you are don't need computing power at a specific time -
send your data to process overnight - and reap the price
benefits.
2. Compute-saving Software
If you pay per compute-cycle, you're going to want to buy
software that reduces that hit. This is completely opposite to the
trend in computing in general. Moore's law is
the famous description of the doubling of computer power and the
halving of the cost over time. So, you build software that may
require cutting-edge machines today that will run on the
wristwatches of tomorrow.
Just as the thrifty try to conserve power now, someday they may
be conserving compute-cycles.
3. Centralized Locations
To return to Bezos' starting point: you don't see generators in
breweries anymore. Because we can deliver the service over
distance, we choose to get scale out of the production plants and
then send the electricity.
We already have regions that are known for their data centers
(Virginia and the Eastern US; Dallas/Austin, Texas) - this would
just push it further. There will be fewer larger data centers that
are strategically located (e.g. near a power source - e.g. AISO.net: a solar farm).
It could also become an industry that was comprised of fewer
larger companies. As
Greg Papadopoulos sees it, it might bring back the truth in the
alleged Thomas Watson (IBM head) quote:
I think there is a world market for maybe five computers.
The qualifier is: we're talking about five cloud
computers.
Links