Design Philosophy of the DARPA Internet Protocols

This paper explains how the goals of DARPA shaped the design of the Internet, providing both historical context and a set of examples for other protocol designers to consider. The most important thing to note is the prioritized goals of the Internet - providing a way to interconnect networks first and foremost, but then, in order of priority, providing fault-tolerance, flexibility in service types and network types, management, low cost, etc. These goals led to the policy of maintaining state at the endpoints, the best-effort, self-routing datagram model, the layering of TCP as one of several possible transport protocols over IP, the byte-stream nature of TCP, etc. It is also interesting to read about some of the design debates that occurred about protocols that we now take for granted. For example, worries about single-character packets from remote login applications flooding the network or about how to manage buffers when dealing with EOL's in TCP seem bizarre today, but they were real concerns at the time, and thankfully led to design decisions that still make sense today (such as allowing fragmentation and reassembly in IP, or removing EOL's from TCP).

The paper is also good at pointing out some of the shortcomings of the TCP/IP model, such as poor support for resource management and accountability. Resource management has been the topic of considerable research efforts in congestion control and QOS. Accountability is coming into the spotlight as denial-of-service attacks begin to be a serious threat, and work such as the Accountable Internet Protocol is beginning to try to address it. Interestingly, Clark suggests some of the techniques used today for these features, such as soft state and keeping track of flows.

One concern that the paper surprisingly does not mention is security, not just at the level of preventing denial-of-service, but also in routing mechanisms, TCP, DNS, etc. There have been numerous attacks discovered and fixed for these basic services. However, in the original Internet, the threat model was clearly just physical destruction of particular hosts and links, and perhaps this was still true in 1988 when Clark wrote this paper. Perhaps this means that DARPA thought the Internet would only be used by themselves, but today it is important to be able to "defend" the Internet in general for the benefit of the entire economy.

In summary, this is an interesting paper to read as an introduction to networking, because it provides historic context, shows the range of problems that had to be addressed, and encourages thinking about the design decisions in the basic protocols of the Internet that we mostly take for granted.