The hidden force behind Google's dominance of search and advertising has always been their back-end server farms. The company has been historically reluctant to discuss in much detail the massive grid engine behind its global services. We know that it makes extensive use of proprietary tools to enable thousands (perhaps millions, now) of commodity, off the shelf servers serve as a single, massively parallel processing system with gobs and gobs of redundant power and storage, all geographically distributed in tens or hundreds of data centers around the world. We know a little bit about some of the tools that run on this massive network, the Google File System (GFS) and Bigtable distributed storage system, from various scholarly articles which have appeared over the years. And we know it works: go pull up Google and run a search right now. When's the last time that has failed for you, in hundreds of individual attempts a day? It's the closest thing on the web to electricity; it's always on unless your wires have come down.

Today, Google is bringing that massive strength to bear on the cloud computing market with the introduction of their new App Engine service. I've been waiting on this for a while now. It seemed obvious that, considering the price/performance advantage that Google's backend enjoys (see this Baseline article for an excellent summation of what is publically known about Google's setup, circa 2006). IN fact, it's that price/performance ratio that is the company's real strength. You won't see pundits debating it, probably because the company doesn't discuss it nor do many of their competitors, but at the end of the day, businesses are profitable when they are efficient, and that efficiency is what may allow Google to clean up in the cloud computing market. While Amazon's EC2 costs pennies per computing cycle, Google has figured out they can give theirs away for free.

In the typical Google way, there are limitations on this initial, inevitably "beta" roll-out: initial sign-ups are limited to 10,000 developers, and there is a bandwidth and persistent storage cap in place (the free only goes so far). But as with all other things Google, you can expect those things to expand and improve, and while some incremental charge will probably be introduced to remove the caps in the system, it's likely that basic use will expand and remain free. While the pundits will no doubt discuss how the service itself will force Microsoft to react, and may impact Amazon's EC2, 3Tera's AppLogic, and other competitors, I think the real story revolves around the computing efficiency that allows Google to so significantly undercut all those companies and others. Bert Armijo, of 3Tera, responded (in my article linked above) to the thought that Google might eventually enter this market this way:

This isn't a unique phenomenon.

Lots of folks built operating systems before Microsoft and Linus, but most of them were large companies selling computers and saw the OS as a tool rather than a product. As such, they built something specific to that task. More recently, the big players didn't jump into virtualization until VMware had proven the market because at the time they'd rather you just bought another server and/or another license.

Back to utility computing, consider Amazon's EC2. They've been pretty open about the fact that it was built to run Amazon and that's evident in it's feature set. If you started building a commercial utility computing system from scratch you'd probably consider it a given that it must run databases, have predictable performance, and allow installing code directly on the system - but EC2 provides none of these. I'm not trying to disparage Amazon, I think they've done a pretty good job, but if you read their forums you'll see their customers have been asking these questions since the service first came out more than a year ago.

More advanced requirements would never be needed, or even desired, by larger players. For instance we work with multiple hosting providers in several countries to give users a choice of data centers. To make it possible to use multiple providers or even switch, AppLogic allows full portability of applications (no matter how complex) with a single command.

We'll have to see if he was right, whether the details of Google's offering handicap it somehow, or whether they can leverage that price/performance ratio to dominate the cloud computing market as well.