Tanks in the Cloud | Hacker News Reader

Tanks in the Cloud

economist.com

26 pointszwilliamson5 comments

The Economist goes to a lot of effort to describe something that is not particularly relevant to getting a sense of the scale of AWS - these are "virtual" instances, which may be spun up and removed fairly rapidly. All we really get is a sense of the transactional volume of instance creation. This is further confused by the roll out of new EC2 products - it's well within the realm of possibility for an individual to roll out 1000 Micro EC2 AMI instances in a single day when experimenting.

In particular, The Economist fails to justify the statement "The results suggest that Amazon’s cloud is a bigger business than previously thought."

More interesting to me, would be the number of _physical_ servers, and their characteristics, in the Data Centers. Or, probably as useful, is the number of "Instance-Hours" along with the scale of those instances.

The most interesting part of the article, for me, was my first (that I can recall - Google Trends shows it's been around for two years) exposure to the phrase IaaS (Infrastructure as a Service). I guess it makes sense that if we commonly use SaaS (Software...), and, in the last year, PaaS (Platform...), that we should bring it back full circle.

I do take exception, BTW, with the phrase "The most interesting layer—the only one that really deserves to be called “cloud computing”, say purists—is “infrastructure as a service”". I was present at the birth of Loudcloud in September of 1999, one of the first (alas for my Stock Options, failed) attempts at a "Cloud Platform" - and I assure you that we had no intention, at the time, of providing the 'infrastructure' as a service. It was all about the Databases and Applications servers.

I worked up some of the original numbers for Cloudkick, and while it doesn't look like it made it into the article, we did try to estimate the actual number of machines running on EC2 (we mostly stuck to US-East) at any one time. In short, based on the incrementing IDs we knew the total number of machines that had been launched. By looking at EC2 machines we've monitored at Cloudkick (at the time, a good portion of the well over 1M servers we had seen were on EC2) we were able to get an idea of what portion of all machines launched on EC2 are still running.

I don't recall the actual numbers off the top of my head, and surely there is some selection bias in servers Cloudkick monitors, but we eventually worked up some estimated revenue numbers and they were right in there with the $500m-$750m (independent) estimates mentioned in the article, so at least we seemed to be in the ballpark.

There's some relevant evidence at https://forums.aws.amazon.com/ann.jspa?annID=877 but people are obsessed with instance IDs to the exclusion of anything else.

Summary for 2010: all SaaS has sales of $11.7 billion; PaaS (I think they mean services like SalesForce, AppEngine, and Azure where you write your apps on their service) $311 million; Amazon $500 million.

Add it all together and you have roughly the size of the remaining U.S. steel industry which is rapidly giving ground to China. My point being the cloud is going to be a sorry economic substitute for dismantling the U.S. manufacturing base.

Original source for most of this is at

http://www.jackofallclouds.com/2010/12/recounting-ec2/

Reply on news.ycombinator.com