December 15, 2002
Quality of Service
[This is a draft of a column that will run in DarwinMag on Wednesday, probably.]
Who could object to quality?
Actually, you do. You object to quality every time you buy a worse-but-cheaper brand, order from the top of the wine list instead of the bottom, or reject the “pro” version of some over-featured piece of software.
And you also object to quality of service every time you get righteously annoyed at the way the waiters fawn over the well-dressed couple while barely remembering to clean up your spilled root beer. And, most important, you object to quality of service when you’re waiting on line and some perceived VIPs are served ahead of you because their quality of service degrades your quality of service.
That’s at the heart of the current debate over Quality of Service (QoS) as the Internet begins to subsume other networks. In fact, David Isenberg (who saw a draft of this column) expands the analog: “Maybe you’re pissed that they got in with no delay while you have to wait (delay). Maybe you’re pissed because they can count on getting in right away, while you never know how long you might have to wait (jitter). And maybe you never get in, thanks to those VIP a**holes filling up the place (lost packets).” One person’s Quality of Service can be another person’s Degradation of Service.
That’s what could be. As it stands, the Internet treats all bits alike. When a packet arrives at a router, it gets sent along regardless of what type of data it encodes. In fact, there’s no way to tell what type of data it’s carrying. Medical X-Ray Bits are moved just as quickly as Pamela Anderson Bits. And it doesn’t matter what order the packets arrive in: if the packets encoding Pam’s eyes show up after the ones encoding her painted toenails (assuming that it’s a photo of her vertical and rightside up), they will be put into the right order by the machine receiving them.
But many of the companies who are using the Internet as a transport for telephone calls say that their bits are different. If you’re reciting the Pledge of Allegiance over the phone (soon to be a requirement, by the way), if the “to the flag” ” bits arrive after the “under God,” your message will be unintelligible. Therefore, telephone bits deserve special treatment. Or so the argument goes.
Some very serious people disagree. Their arguments are of three types.
First, QoS is impractical. There are indeed bits in IP packets designed to indicate “type of service,” but no one uses them. There’s not even agreement how to interpret them, much less how to rank them. More important, the Internet routers would have to be set to act on thost bits which would require a massive retooling of the Net’s “operating system.”
Second, QoS is the wrong solution. According to this line of thought, QoS is only required if there’s a scarcity of bits available. It’d be far better to solve the QoS issue by opening up the sluices of connectivity: light up the “dark” (unused) fiber, open up the spectrum for public access, install more powerful routers, get with the IPv6 program. With sufficient bits and sufficient throughput, voice packets will arrive in time without having to always arrive first.
Third. QoS violates the principle of the Internet’s architecture. The Net has succeeded precisely because it does nothing but move bits from A to B. This is the “ End-to-End” theory described by Saltzer, Reed and Clark and the “Stupid Network” as described by Isenberg. Part of the simplicity that keeps the Internet humming is the fact that it treats all bits alike. Further, the fact that the Internet is not optimized for any particular applications means that it is optimized for innovation; “tune” the Internet for the VIP du jour and you will de-tune it for other applications. Let the telphone guys and gals change the way the Net works so that it’s better for voice and then tomorrow change it so that it works better for broadcasting TV shows, and soon the other types of bits we care about will have trouble getting through the line at the check-out counter.
So, even though in theory, we could provide QoS in the short term and open up connectivity in the longer term, doing so would mean obfuscating the true strength of the Internet: It’s no Strom Thurmond when it comes to bits.
Resources
Glenn Fleishman just published a helpful weblog entry on the topic.
Lawrence Lessig just posted a terrific article to Dave Farber’s mailing list, as did Karl Auerbach and Bob Frankston.
David Isenberg’s newsletter in 1999 ran a clear explanation of the issues. The article is based in part on research by Andrew Odlyzko; there’s a long interview with Odlyzko here.
This column resulted from my participating in (well, auditing) an email thread among Glenn Fleishman, Bob Frankston and David Reed. Obviously, they are not responsible for my stupidity, carelessness and poor personal hygiene.
