Friday, November 12, 2010

The Cloud Scalability Distraction

I have this sneaking suspicion that Amazon has done something extremely clever with the cloud. When they launched their cloud services a few years ago, they made a big deal about how well they scaled up - S3 can store unlimited data! You can start up as many EC2 instances as you want! They even drive the point home with their pricing - S3 pricing is tiered, and there's a tier for people storing more than 5 petabytes of data, "proving" that S3 will easily scale to that amount. (Having that tier listed on the website is a bit unnecessary - if you're going to be spending millions of dollars per year on S3, you're probably going to get to talk to a salesperson face-to-face, just saying.)

Why is this clever? Because all of Amazon's competitors in the cloud space followed suit, and talked about how well their services scaled up, too. Meanwhile, Amazon's been advancing in precisely the other direction - the ability to scale down. For example, they recently launched "micro instances", which are cloud servers that rival the cheapest VPS providers in price.

Why does this matter? Because scaled-down cloud services are going to be the next revolution in computing. Right now, "cloud" is more of a marketing victory than anything else; there's nothing there that a few competent sysadmins and devs couldn't put together in a week or two, for a few times the price. The biggest advantage is the pricing (and it's telling that Amazon is on the forefront of it; the expertise they had that contributed most to the cloud wasn't technology, it was payment processing on a massive scale). Cloud services are still pretty coarse-grained, compared to what they could be.

Right now, the technology that would make the cloud "revolutionary" hasn't been invented yet. (I may have said this before.) I think that that technology is going to be something that enables partitioning up cloud services a few orders of magnitude more efficiently than we can today. Imagine a system where the resources can be sliced finely (and cheaply) enough that it makes sense to integrate them into desktop applications, and have the user pay for it - that's when you'll really have something revolutionary on your hands.

(aside: the title of this post makes me think of the big bang theory T___T)

No comments: