Thursday, January 31, 2008

Amazon Elastic Compute Cloud: Heavy Use

A growing computing architectural theme is the move of functions out of proprietary data centers and "into the cloud," a return in some ways to the days of time sharing as a computing architecture. So it is that 330,000 or so developers have registered to use Amazon Web Services, up more than 30,000 from the prior quarter.

And those users are driving traffic and compute cycles. Amazon Elastic Compute Cloud (EC2) and Amazon Simple Storage Service (S3) consumed more bandwidth in fourth quarter 2007 than was consumed in the same period by all of Amazon.com's global Web sites combined.

At some point, the availability of cloud computing resources is going to fundamentally alter the tradtional "build versus buy" equation that has had enterprises and other large entities building and maintaining their own data centers. At some point the computing framework used by smaller entities and individuals is going to change as well.

At some point, one has to wonder whether communications and computing, increasingly intertwined, might also be thought about in different ways.

To the extent that servers, air conditioning, power, space and communications are the underpinning for applications, and to the extent that enterprises and individuals typically only care about infrastructure to the extent that it enables use of applications, one is lead to ponder the notion of outsourcing of infrastructure.

To what extent must even a large provider "own" its own conduits, routes, physical media, servers and software of an infrastructure sort? To what extent can those things be sourced more extensively on a "buy" basis rather than a "build" basis? In how many more use cases will it make sense to source wholesale capabilities from other providers instead of building, owning and operating facilities?

To the extent that it is the "computing" that matters, not the "computers," one also might ask whether it is the "communications" rather than the "network" that matters.

No comments:

Costs of Creating Machine Learning Models is Up Sharply

With the caveat that we must be careful about making linear extrapolations into the future, training costs of state-of-the-art AI models hav...