The Design Philosophy of the DARPA Internet Protocols

The paper explains some of the motivations and justifications about the design and implementation decisions of the Internet and its well known protocols. Perhaps the most gratifying part of the paper is the discussion about the necessity to separate TCP into IP and eventually add UDP (at least this bit of history clears up the oddity in the TCP versus UDP naming). Of course, this was gratifying because it reinforced the conclusions made in the paper "End-to-End Arguments in System Design".

In connecting these two papers further, I found the statement "a surprising observation about the control of variation in delay is that the most serious source of delay in networks is the mechanism to provide reliable delivery" to be most poignant. That is, it highlights the correctness versus performance trade-off that I discussed in my previous post. In this case it argues that performance may actually suffer from reliable delivery (where as before we were claiming we could/should use it for performance).