UNIX System Administration Handbook - Evi Nemeth [159]
UDP is a packet-oriented service. It’s analogous to sending a letter through the post office. It does not provide two-way connections and does not have any form of congestion control.
TCP is a polite protocol that forces competing users to share bandwidth and generally behave in ways that are good for the productivity of the overall network. UDP, on the other hand, blasts packets out as fast as it can. Better behavior for UDP is being worked on, but it doesn’t yet seem to have become a high priority.
As the Internet becomes more popular and more crowded, we need the traffic to be mostly TCP to avoid congestion and effectively share the available bandwidth. Measurements of protocol utilization over the past few years show UDP traffic increasing from about 5% of the bytes in 1997-98 to about 7% in 1999-00. Applications such as games, music, voice, and video are starting to make their presence felt.
This chapter introduces the TCP/IP protocols in the political and technical context of the Internet. Unfortunately, even basic networking is too big a topic to be covered in a single chapter. Other network-related chapters in this book include Chapter 14, Routing, Chapter 20, Network Management and Debugging, and Chapter 16, The Domain Name System.
The next few sections include background material on the protocols and politics of the Internet and are quite opinionated and fluffy. Skip ahead to page 271 to go directly to the gory details of IP, or to page 306 to jump to vendor-specific configuration information.
13.1 TCP/IP AND THE INTERNET
TCP/IP and the Internet share a history that goes back several decades. The technical success of the Internet is due largely to the elegant and flexible design of TCP/IP. In turn, the leverage provided by the Internet has helped TCP/IP prevail over several competing protocol suites that were favored at one time or another for political or commercial reasons.
A brief history lesson
Contrary to popular belief, the Internet is not a Microsoft product that debuted in 1995, nor is it the creation of a U.S. vice president. The progenitor of the modern Internet was a network called ARPANET that was established in 1969 by DARPA (Defense Advanced Research Project Agency), the R&D arm of the U.S. Department of Defense. The ARPANET eventually became the NSFNET backbone, which connected supercomputer sites and regional networks.
By the end of the 1980s, the network was no longer a research project and it was time for the National Science Foundation to extract itself from the networking business. We transitioned to the commercial Internet over a period of several years; the NSFNET was turned off in April of 1994. Today’s backbone Internet is a collection of private networks owned by Internet service providers (ISPs) that interconnect at many so-called peering points.
In the mid-1980s, the Internet essentially consisted of the original ARPANET sites and a handful of universities with Digital Equipment Corporation’s VAX computers running Berkeley UNIX on 10 Mb/s Ethernets connected by 56 Kb/s leased digital telephone lines. Every September, when students came back to school, the Internet would suffer what became known as congestion collapse. Van Jacobson, a researcher in the Network Research Group at Lawrence Berkeley Labs, would look at the protocols’ behavior under load and “fix” them. The algorithms we now know as slow start, congestion avoidance, fast retransmit, and fast recovery all arose from this context.
Moore’s law (the rule of thumb that hardware speeds double every 18 months) and market pressure have greatly accelerated the development of the net. Since the late 1980s when the current TCP algorithms were stabilized, the speed of network interfaces has increased by a factor of 1,000 (from 6% efficiency on early 10 Mb/s Ethernets