Your Humble Blogger just happened to come across this note, and although I understand almost none of it (and, frankly, am content with that), I did come away with a new respect for the Google enterprise. I’ll see if I can explain why.
It seems that many companies quite properly add the cost of maintaining the system to the actual hardware cost when making hardware investments. Actually, Your Humble Blogger wouldn’t be at all surprised if many don’t, but there it is. Anyway, since having a lot of cheap servers is much more expensive to maintain than a few large durable servers, companies such as those Rich Skrenta describes having previously worked for make the cost-efficient decision.
Except ... that assumes that many cheap servers were in fact really expensive to maintain. If I understand the story correctly, Google found a way to make a very large network of cheap servers reasonable to maintain. Having done that, they set themselves up with a lot of servers. Like, a hundred thousand.
Now, here’s the clever bit. If you are interested in embiggening your operation, finding a cheap way to maintain a lot of servers is a very good idea. Google was able to go from 50,000 servers to 100,000 servers fairly quickly, and with a, well, absurdly large investment, I’m sure, but commensurate with their size. And that’s because they already have the experience of going from 10,000 to 50,000 servers. It means that if Google wants to get into the email business, they can, because they have a hundred thousand servers, and lots of experience maintaining them. It also means, and this is the clever part, that if somebody is interested in trying out some spiffy new code that would be slightly better and really computer-intensive in what Mr. Skrenta calls CPU seconds, they have the capacity to try it out, and if people like it, they can do it.
In other words, because of the way they chose to set up their business, it costs them less to improve their widgets than it does their competitors. Not necessarily to make their widgets, but to improve them. But if people want better widgets, then pretty soon it will cost you less to make those new, improved super-widgets now with extra widgetosity than it will your competitors. And then you win.
When setting up a process, how many companies really do a good job of making it easier to improve their stuff later? No wonder Michael Porter left a note, even if it wasn’t the real Michael Porter.
Redintegro Iraq,
-Vardibidian.