For a substantial part of its history, the Paris Metro system used a novel pricing scheme. A first-class ticket cost twice as much as a second-class ticket, but entitled you to sit in the first-class car. And what made the first-class car sufficiently luxurious that you'd pay double to ride in it? The big number "1" painted on the side, and nothing more. The first-class cars and second-class cars were identical in every important respect, save only that only holders of first-class tickets were allowed to ride in the first-class cars.

This statement is not as vacuous as it may seem. Consider the experience from a rider's point of view. Those with no particular preference rode second-class and saved their centimes. Those seeking a quieter, less crowded ride paid the extra fare and rode first-class. The system was self-regulating, in that as crowding in second-class increased, the temptation to take the expensive seats also increased, thereby relieving some of the pressure on the second-class cars. Economically, we're accustomed to thinking of product differentiation as a cause and price differentiation as an effect. Under the right circumstances, though, the relationship takes on more of a chicken-and-egg quality: the Paris Metro system managed to automagically transmute a price differential into a quality differential.

The computer scientist Andrew Odlyzko, noticing this fact, came up with the most elegant quality-of-service proposal I've seen. Trying to build differentiated services and preferential treatment into routers and protocols is a technical and logistical quagmire whose uncertain traversal has driven to the brink of madness many of the best minds in computer networks. Instead, Odlyzko proposes simply to build parallel Internets, ranging from "cheapass" to "Rolls-Royce" distinguished only in their usage costs. It's a beautiful idea: we already know how to build one Internet, so in some sense, it would be much simpler to build serveral Internets than to increase the complexity of the routing algorithms in the one we've already got.

Of course, the idea is completely and utterly unworkable as a serious proposal for the Internet. Its chief merit consists, I think, in revealing, by way of comparison, how far beyond utterly unworkable other quality-of-service schems are. That is to say, present quality guarantees always exist at the sufferance of the the Internet as a whole. We're relying on general excess capacity and the laws of probability to see us through, and if you need something more reliable than the Internet can provide you, you may need to roll your own. Enterprise Rent-a-Car, for example, uses its own satellite network to keep its locations connected to the central computer systems. Other than not being at the mercy of Star Wars Episode I trailer-induced bottlenecks, Enterprise also enjoys a certain (partial) immunity to script kiddies as a consequence.

I've gotten started on this train of thought because of the recent sense of crisis among many free and formerly-free online services (and also many former online services, to be precise). The problem of finding a revenue model that can bring in enough money merely to pay the marginal costs of the service has plagued all sorts of enterprises. Many have had to embark on crash programs to find features they can cram into a "pay" version that will somehow be able to keep the whole's head above water. The Paris Metro Pricing scheme may not be strictly applicable, but I think its basic insight is worth keeping in mind: the very act of differential pricing, if properly applied, can provide just as effective a filter on customer behavior as any "genuine" difference in the relative quality of your products.

The infamous Blogger Server Fund wasn't just about community spirit, and it wasn't just about the servers themselves. Think of it this way: Blogger positioned itself, quite publicly, as a service associated with voluntary donations. No compulsion, sure, but their splash page made certain to let you know about the server fund. Mmmm, low-level guilt trip. The result: regular Blogger is now, very slightly, a "premuim" service. Much less so than No-Branding Blogger, sure, but still with certain stability guarantees that always-and-forever absolutely free sites don't provide. Yes, you could attribute that stability entirely to the server fund itself, but I think that attribution would be wrong. "Cheaper" alternatives hit their capacity and collapse under the load -- both network and psychic -- of users making the sorts of demands on the system users tend to make. Blogger probably drove some people away, if the loud moral indignation is any indication, but my point is that, in some sense, Blogger made the choice that it didn't really want those users, who were unwilling to accept a little bit of a feeling of responsibility along with their blogging privileges. It's Paris Metro Pricing in action: from many users' perspective, there really is no significant difference between Blogger and its competitors, and yet the system starts shaking them out, some here and some there, and everyone feels that they've come out ahead of those poor suckers who bought the other kind of ticket.