Distributed vs. Utility Computing

(new longer version)

To elaborate on a point I made in my previous post, there seems to be a tension between the ideas in Carr's book about a "utility computing" future and a trend in the opposite direction, towards "distributed computing." Distributed computing is the idea that computing power will appear in more and more places, and make all kinds of devices more powerful or "smarter." Even more so than Carr's predictions, distributed computing is clearly happening already. Our cars are largely run by computers, as I pointed out earlier, my toaster has a microchip. Above all, one word - iPod.

The NY Times had an article yesterday on the trend, if you want to read more.

The point is this - is the distributed computing trend pushing against the forces Carr identifies as moving us toward utility computing, or can the two trends coexist? Coexistence is possible - imagine that some computing power is distributed to small devices in homes and businesses, putting it right where it needs to be used, while the remainder becomes provided over a distance by utilities. The big change from the current world is that the device between these two poles - the PC - disappears. Small jobs get performed at the point of action, and big jobs get outsourced. My guess is that this would be Carr's prediction.

I'm not sure it's right, however. Moore's Law makes computing power so cheap that even the chips in our toasters will be quite capable. It makes little sense to install a weak chip when, for essentially the same price, manufacturers can install a chip that outstrips a personal computer of only a few years ago. We may in the near future have a data center's worth of computing power in our homes, simply distributed across all of our devices (and the definition of device will expand - think smart clothes). With that much power in the home and in businesses, it will probably be much cheaper to harness it than to supply computing from a distance. The app that kills utility computing may be the guy that can get you ski jacket to talk to your microwave and work together to sort cells in a spreadsheet (or, of course, find aliens).

Essentially, this debate comes down to the relationship between bandwidth costs and processing costs. Carr's point depends on the economies of scale from centralization of processing exceeding the costs of distribution. My intuition is that processing is and will continue to be cheaper than transmission. I share Mr. James' skepticism about long-term decline in bandwidth costs, while we know for sure that the rate of increase in processing power per dollar is rapid and accelerating (Moore's Law). Carr's analogy to electricity (in my opinion, the best idea in the book so far, though it may not be original) is useful here - it makes sense to generate electricity centrally because the savings from doing so exceed the costs of transmitting it widely. This might not always be true - if cheap home solar power, fuel cells, cold fusion, etc. become feasible, it might be cheaper to generate power at home. A few people do this already, even contributing power back to the grid and being paid for it - though it doesn't make economic sense on a large scale. Just as cheap electricity generation technology would make electric utilities obsolete, cheap processing power might render IT utilities a stillborn idea.

A deeper analysis would go beyond these two inputs (generation and distribution costs) and look at other costs - storage and labor, in particular come to mind. Would economies of scale reduce these costs enough to counterbalance distribution costs, even if they do not reduce processing power costs? Maybe. It's an empirical question. It seems likely that labor costs would be a great savings - Mr. Bednarz' point in class that labor might be the greatest savings from centralization seems intuitively convincing. I would guess that storage tends to play a similar role to that of processing power - storage costs are decreasing at a similar rate - iPod storage size has increased even more rapidly than processing power.

In short, there are lots of inputs into the equation that determines the structure of the IT industry. They seem to cut in different directions, which both makes prediction difficult (leading me to question Carr's conviction that his model is right) and means that a variety of models are likely to exist. Change will occur, but it is possible for it to occur in many ways simultaneously - all prophets might be partly right and partly wrong. This makes me less worried about second-order issues like privacy (opt-out will be possible without being cut off from the world entirely) and natural monopoly (changing and coexisting models make dominance harder to achieve and less pervasive when it occurs).

Comments

[Ruben Rodrigues]

In my opinion: Yes, we can all just get along.