Saturday, February 5, 2005

The Third Wave

Originally published February 5, 2005

We're almost at the tipping point when the tried-and-true circuit switched telephone network gets replaced by packet technologies. While there are a few significant technical hurdles to be worked out, it could be the regulatory issues that get in the way. That's another way of saying that the folks in the telecom world who currently have all the power and money have not yet been satisfied that they will remain in power after the technological change.

Almost all of the technologies issues have been solved except one: variable packet latency. Here's the best way I've come up with describe what this means...

"Movies" on film are captured as a sequence of still frames. Film is typically shot at 24 frames per second (fps). As long as the projector showing the film plays it back at 24 fps, we accept it as a continuously moving image. But if the projector slowed down, or stopped for just a fraction of a second, it would break the illusion.

The same thing goes on with digital music devices, such as a CD player or an iPod. The music is sampled and stored in digital frames, and must be played back at the original sampling rate to sound natural. We have probably all experienced a CD skipping (in spite of read-ahead caching and error retry logic).

The circuit switched telephone network works much like the pathway between the music recording studio and your iPod. The sounds entering the phone network are sampled (at 8KHz) and converted to digital packets that are sent to the other end of the call. There it is converted back into analog sound so you can hear. One of the design criteria for the telephone network was that these 8KHz x 8 bit samples (hence the 64Kbps 'bearer' channel) get pushed through the network with a minimum of delay, and no varience in the interval between sample arrival times. Because of this, voice conversations through the telephone network has achieved ever increasing fidelity (e.g. the Sprint "Pin Drop" ad campaign).

But that audio quality has a cost. A wireline telephone call has 64Kbps of network capacity assigned in each direction (full duplex), even if neither party is speaking. There are some techniques to lessen the waste, but they can cause degregation in the audio quality. When you rip a CD to MP3 files, you make the same kinds of choices: high fidelity and big file, or a small file and less fidelity.

The issue with voice calls over a packet network isn't so much about the bandwidth required as it is the variability of packet arrival times. After all, many people have internet connections in their homes that exceed 100Kbps, much more than is needed to maintain circuit switched quality. But herein lies the fundamental difference between a circuit switched network and a packet network. As noted above, a circuit switched network guarantees constant packet arrival times by reserving 100% of the capacity needed for the duration of the call. A packet network consumes bandwidth only when it has a packet to transmit/forward. Each switch in the packet network waits for a packet to arrive, makes forwarding decisions, and retransmits the packet. This takes time, and the amount of time it takes can vary based on things like the number of other packets that it has to deal with at that moment in time.

When the packets are full of computer-to-computer communications, they can be delayed or even lost, and the software on each each (eg the TCP in TCP/IP) can recover without any loss. But if those packets contain sampled voice from a telephone call, variances and losses are like bad splices in that movie that's been shown 1,000 times. It doesn't take very much of this to be annoying.

But this problem will be solved. Quality of Sevice (QoS) protocols and algorithms along with faster switches and transmission media will be applied so that the latency variability can be kept below our threshhold of detectability.
The biggest obstacle will be about the power and money.

The breakup of AT&T and the emergence of competitive telephone companies started an evolutionary change (I'll write about evolution in another entry) that has not yet completely played out. Economists talk about "natural monopolies" when describing industries in which the cost of the infrastructure is so expensive that once a single company makes the investment, there is no economic reason for additional companies to jump into the fray. The traditional examples are utilities like the water/sewer services and the telephone service. In many countries, these agencies are operated by the national government.

When MCI won the regulatory battle to gain permission to compete with AT&T for long distance service, it invalidated the argument that telephone service was a natural monopoly, and eventually led to decision to break up AT&T. But there was a nasty problem to be solved. In exchange for its monopoly status AT&T was ordered to sell its services at "cost + a reasonable profit", and it was up to the regulators to approve the rates AT&T could charge. But AT&T argued, reasonably, that it cost a lot of money to run telephones out to the rural areas of the country, and if they charged those people the true cost, none would have a phone. So the FCC (which was set up on the model of the ICC, which was created to reign in the railroad barons) said AT&T could charge a premium for long distance service, which was viewed as a luxury, and use that premium to subsidize the cost of rural telephone service.

In the breakup of AT&T, the long distance service was separated from the local service company. To make sure the local companies still had the money to provide local service in rural areas, the FCC allowed the local companies to charge a "common carrier access fee" to the long distance companies. This was the mechanism used to transfer the long distance premium to the local phone companies.

When the data networks, like Tymnet, Telenet and CompuServe, started showing up in the early 1980s, the local companies felt they should be treated like a long-distance carrier and be required to pay the same access fee. Arguments were made to the FCC, led by CompuServe, and an exception was granted for data networks that remains in force.

As long as the data network carries only data, the phone companies could begrudgenly tolerate this. But with Voice over IP (VoIP) technology quickly gaining ground, the local companies again have a reasonable argument for ending the exemption. The trouble is that the telephone industry is reconsolidating.

When a regulated industry is cut loose to full competition, it seems a predictable cycle takes place. First there are many many little startups which appear to compete for a piece of the pie. Most don't make it, and the industry settles down to maybe ten players who are viable at the existing price levels. A problem with these capital intensive industries is that they have high fixed costs, but yet must maintain some growth capacity to take away market share from the others. The more companies who are competing, the more aggregate surplus there is in the industry. That surplus leads to price competition as the ten companies try to fill their "empty seats" -- as is the case with the airline industry. Only a few of those companies will have the capital available to survive a protracted price war, and the industy will probably resolve down to an oligopoly of three very large players. It happened that way with the auto industry (which was never regulated but is definitely capital intensive), and you can see it heading that way with the airlines.

AT&T is gone except as a brand name. AT&T Wireless is owned by Cingular (which is owned by SBC and BellSouth), and the rest of AT&T was just purchased by SBC. Quest, or someone else, is likely to buy MCI, which is what's left of the old MCI and Worldcom. All this consolidation is confusing the long-distance, local, and internet economics, and therefore the degree to which each component should be regulated. Do land-line common carrier access charges make any sense when many people in rural areas can get VoIP service via their cable TV vendor?
One would have to predict that we entering a time when local phone companies, long distance companies, and cable TV companies all get thrown into a big Cuisinart, and then poured out into three viable companies. Anyone want to guess what the names will be?

No comments: