But first, a rant about silver bullets. CEOs love them. Shareholders love them. Politicians love them. It’s too bad they are hardly ever real. The world has a way of keeping itself in balance by using opposing forces and feedback loops (you know, like our government used to). Sure, we find disruptive advancements in technology, math, and even anthropology, but even these usually have some sort of cost or downside. While you keep trying to make alchemy work, we’ll just keep on finding significant, but incremental, improvements, and we’ll see who wins.
Back to our main story. The Australian site IT News posted this article that’s been linked to about 10% of the geek websites in the InterTubes (did I spell that right?) on a 7-month study of availability and performance metrics for the big cloud vendors: Amazon EC2, Google AppEngine and Microsoft Azure. The short version of this study is that the response time of all of them were highly variable and unpredictable, though there was often correlation to the time of day, and the reliability was less than they would have expected.
This is pretty much what I would expect. All of those companies offer many services for freee, but this is the money-making end, and just like all those lowballing ISPs that went out of business, you can’t offer services that cost you money with high quality and low price for very long. And smartly, these companies didn’t try. Borrowing computer time from someone else and running jobs over the internet is simply not going to perform as well as running things locally. The critical difference, though, is with cloud computing, you can’t really do anything about it. It’s out of your hands.
I worked for a company once that did huge analytics run in a Beowulf cluster with 43 nodes. Each node costs under $2000, and when they got a little extra money, they would buy more nodes. That’s how they got up to 43 nodes; not leveraging the company out to a VC for the money. And it worked damn well.
But now it’s time for another rant. Sorry; that’s just how I am. As one commenter to the article pointed out, this article is surprisingly light on details. OK, it’s in an IT magazine, not an academic journal. But every single reference to this study I found linked back to this article as ground zero. Nothing else gave a link to the original study, or any other source with more information. In fact, The University’s own website, and the author’s bio page don’t even have links to it. I have my doubts about the lineage of this study, though I certainly believe the conclusions.