Have you ever thought about what happens behind the scenes when a telco decides to build a new network?
It runs something like this. Their market research folks go out and ask customers what sort of communications they will be performing, and how often. They come back with these vague labels (“surf the web”, “chat”), and the network folks guess how many bytes or minutes each interaction uses, and how many interactions there are. They then go off and scout around for a network technology that maximizes the overall throughput (bytes per hertz, switches per second, etc.). The network gets built, marketing set a price, and the punters roll in.
Now, there could be more or fewer customers than the business plan anticipated. They could be more or less active than planned. Their communications could be lighter or heavier than planned. If there are more bytes and minutes than the system capacity, then a whole new truck roll needs to happen to expand the capacity of the network.
So in short, the price gets fixed, and the capacity then varies according to demand.
But this isn’t the only way to approach the problem. What would happen if we turned the whole thing inside out? What if we fixed the capacity, and then varied the price? Whooaaa! What would that even look like?
In one sense, we’re already there. Your Internet access provider has a big pipe upstream. You can buy yourself a local pipe, with a nominal maximum speed. A menu of speeds and prices are normally on offer. But all it really gives you are more chances to sling a packet in the direction of the upstream pipe. Statistically, you do better than your cheap neighbor who bought a thinner pipe. You aren’t buying bandwidth, but rather priority as a emergent property of a probability distribution curve. Of course, this faux priority rationing is just a short-term effect: if the upstream service is lousy enough, and you have a choice, you may leave for another access provider.
Another way of doing it is through traditional QoS. Now, what you don’t need is to fiddle with packets on a per-service basis. This approach has been thoroughly debunked, at least with respect to wireline networks. It’s so easy to add more capacity to wireline services, it just isn’t worth building more silicon to ration stuff out. Repeat after me: optimizing for one application de-optimizes for all others (Om Mane Padme Hum, etc.). Only your end point should decide the priority of packets between services.
On wireless networks, the world of abundance has yet to arrive. Notwithstanding the recent theoretical research that wireless networks have infinite capacity, we’re stuck today with wireless technologies that simulate wireline channels. And the pipes are really thin. Demand can easily exceed supply. Adding capacity is slow and expensive. Even with wide-area OFDM “4G” networks, this isn’t going to change soon.
Who decides which people get priority access? Delegating priority to the end points only works to the extent that each end point can express relative priorities for its own communications. But everyone wants to be first in the queue when the collective bandwidth needs to be divided up! Thus priority of access is a different concept to priority of service. The difference is important.
Imagine a future where your laptop doesn’t just come with spiffy new Centrino wi-fi hardware. Instead, if embeds a wide area wireless network card, along with pre-provisioned nationwide low-priority access. The cost of lifetime service is built into the device. The huge volumes keep incremental prices low. Of course everyone knows that accident, obsolescence and changing fashion will guarantee a steady stream of replacement hardware business.
Perhaps the network capacity is gently scaled to ensure the great majority of users enjoy a dial-up level of connectivity in the great majority of cases. As appetites are whetted for always-on wireless, you up-sell higher access speed (or lower latency — a discussion for another day…). Sometimes at quiet periods even the cheapskates will get to experience great throughput. You under-promise and over-deliver.
So you don’t need crazy Enronesque dynamic pricing to ration fluctuating network demand. Static priority pricing schemes like those above are also an option. This fix-capacity-vary-price makes the economics align more with the reality of the physical network. It makes for more stable business cases. It expands the potential market to new low-end applications. It enables a crude form of price disrimination, because people wanting to run low-latency high-bandwidth applications with great reliability will need to outbid others riding on the same pipe.
In many ways it closely resembles the way airlines work: they don’t lay on an extra plane just because the flight has sold out. Every seat that flies unfilled is a lost revenue opportunity. On a network, every wasted hertz, electon or photon represents wasted capacity that can never be re-used. You can’t store bandwidth. It’s a perisable commodity with a lifespan of zero. Price it to sell!
Could it work? Time will tell…
Posted by Martin Geddes at 03:43 PMTrackBack URL for this entry:
http://www.telepocalypse.net/cgi-sys/cgiwrap/mgeddes/MT/mt-tb.cgi/56.
Actually, unless your ISP is doing fancy traffic shaping with differentiated classes of service (hint: nobody does), whether you have a DSL or a dialup connection has no direct bearing on how much bandwidth you will get. If the ISP is so badly oversubscribed that each active user has only 20 kbps on average, dialup and DSL users will all have the same bandwidth, 20kbps. The router's probability of dropping your packets (thus giving your TCP connections a signal to throttle down in the face of congestion) is proportional to the share of capacity you are taking. This is enhanced by mechanisms like RED (http://www.magma.ca/~terrim/RedLit.htm). If the average bandwidth is greater than 56K, dialup users will "enjoy" full 56K, and DSL users will have 56K plus a fair share of whatever is left over.
Posted by: at November 10, 2003 07:05 PM