This article explains in detail the design decisions behind the DARPA 
Internet protocol whose motivation has been greatly misunderstood. 
What makes this paper different from other papers is this work focuses 
not on the specifications (HOW the protocols work) but on the 
motivation (WHY the protocol is as it is.) 

The fundamental goal for the Internet architecture is the multiplexed 
utilization of the existing interconnected networks. Therefore, packet 
switching used by the networks was selected over circuit switching as 
a fundamental component of the architecture, and gateways were adopted 
as the interconnecting technique. The second level goals are listed in 
the paper in their order of importance: survivability, varieties of 
types of services and networks, distributed resource management, cost 
effective, scalability and accountability. Because the Internet was 
designed to operate in a military context, survivability in the face 
of failure is the most important goal in the list. The Internet took 
"fate-sharing" as the approach to survivability instead of replication 
to reduce the design complexity and enhance robustness. The goal of 
providing different types of service in the architecture caused TCP 
and IP to be separated into two layers, with TCP providing the 
reliable sequenced data stream and IP providing a basic building 
block. The Internet is particularly successful in incorporating a wide 
variety of networks because of its minimum assumptions about the net 
function. The paper addresses how other design goals in the list 
influenced the Internet architecture, such as distributed management 
(especially in the area of routing), efficiency, cost of attaching a 
host to the Internet and accountability.

This paper also explains the rationale behind the datagram, TCP 
mechanisms and their contributions to the achievement of the above 
goals. For TCP, the reason for acknowledging bytes instead of packets, 
the EOL semantics were given as examples of design decision making.

In the end, the author discusses future improvement in Internet 
architecture design. A better building block than the datagram for the 
next generation of architecture is suggested.