Netheads versus Bellheads is a classic manifesto of the packet-switched world, written by François Ménard and David Isenberg. The amazing thing about it is how well it predicts that the perceived need for predictable Quality of Service (QoS) from the phone companies and their ideological fellow-travellers will clash with the needs of the Internet.
It is the authors' view that, in the PSTN [public switched telephone networks], there is only one application, setting up or tearing down 64 kilobits channels. Since there is, in essence, only one application in the PSTN, it is impossible not to bundle service quality with transmission quality. End-user experience is a direct function of how well the PSTN performs this single application.
So what does QoS mean in an Internet environment, then? The difficulty with the concept of quality of service in an Internet is that it derives from ideas that really have no place there, like driving a 64,000 wagon train down the highway. For example, packet loss is not a degradation of service, it is rather a mechanism to make it possible for multiple applications to share the same finite bandwidth. On a highway, when there is more traffic than the road can handle, cars slow down and sometimes crashes happen. This is usually not a problem as cars simply route around the accident. By contrast, on railways, crashes and derailings are catastrophic.
If all we ever wanted to do was to talk on the telephone (one application), the PSTN would have remained fully adequate. However, as soon as computer to computer communications became important, the rigidities and expense of circuit switched voice networks became apparent. Since every projection of bandwidth requirements shows that data traffic will expand at a minimum of two orders of magnitude faster than voice, the assumptions of best effort engineering currently embedded in the Internet are likely to hold true indefinitely.
(via Jon's Radio)