July 27, 2011
Cost-effective? They keep using that word. I do not think it means what they think it means.
If you read Slashdot yesterday, you might have seen that some researchers from Microsoft released a paper suggesting that you could use servers to heat homes and businesses.
OK, first things first - this is creative thinking, and I think that should be encouraged. "What if..." is a great question, and it leads to cool stuff. So "Yay imagination!"
On the other hand, I find it really hard to believe that they actually got a paper out of the idea. You can read the thing here, and it does actually mention several of the problems I'm going to bring up, but it doesn't even really gloss over them - it just treats them as non-problems.
The big one? Summer. There are very few places where heaters are required year-round. Mostly we spend lots of money removing heat from our residences, not trying to get more in. The paper references climate zones, and it also mentions that a low-footprint seasonal datacenter could be made from old inefficient servers. More on that thought later.
The next big hurdle? Bandwidth...and there are lots of problems to be had here. The authors mention an additional price of installing a T1 to each home that has one of these, and the price increase they give is $2,640 / year. I'm not sure where they get their T1s, but that comes out to $220 per month, which is half of what I got my cheapest T1 for.
You could assume that the residents will be partaking of the T1, too, but I don't know many people who would willingly drop their 10+Mb/s best effort cable modem for a 1.5Mb/s leased line that they're sharing with a bunch of servers...which leads to the next problem. At 1.5Mb/s, what are you doing? You could essentially have a bunch of compute nodes, but you're not going to be serving anything of consequence across that line. And don't even get me started in installation fees.
Then there are other issues that crop up when you take a step back... Presumably, servers of the future will be even more energy efficient than we have now, so the number of servers required to heat a particular space at this moment is smaller than the number that will be required to heat it in 5 years and is probably (hopefully?) vastly smaller than will be needed in 10 (assuming you're not still going to be using the equivalent of Pentium 2 processors).
Also? I have no idea where I would put a bunch of servers. My garage, I suppose? I guess most people could build a shed according to specs, but there's a decent amount of work involved in getting a building that can hold a ton of servers.
Anyway, it's certainly not going to work for the mass market as pitched. I can definitely see where a certain niche of people could take advantage of it, if Data Furnace as a Service (DFaaS) was offered (bet I'm the first Google hit on THAT phrase), but it would be expensive to do it yourself with no one to sell your cycles to, and it would be limited to the most northern and southern climes in areas that have cheap readily-available bandwidth.
If that's you, then maybe you could consider starting a company to do it. It would be kind of fun to spec out and build a bunch off small datacenters like that. And like I said before, it's a neat idea.
What do you think?