This paper presents a design principle end-to-end argument and explains 
its importance in helping guide placement of functions among the 
modules of a distributed computer system. End-to-end argument states 
that the function in question can completely and correctly be 
implemented only with the knowledge and help of the application 
standing at the end points of the communication system. Providing that 
questioned function as a feature of communication system itself is not 
possible. 

There are two reasons to implement certain function end-to-end:
 1) The service provided by the low level subsystem may be shared by a 
number of applications, and those that don't need the service will pay 
for the unnecessary overhead; 
 2) The low-level subsystem may not have as much information as higher 
level to do certain jobs efficiently. 

The authors use many examples to support this point. In the "careful 
file transfer" example, two approaches: "step by step check and retry" 
and "end-to-end check and retry" are carefully compared under different 
assumptions. It is concluded that the "end-to-end check and retry" is 
much more efficient than reinforcing each of the file transfer steps to 
guarantee the correctness and furthermore, placing functions at low 
levels is just an engineering tradeoff based on performance. Other 
examples such as delivery guarantees, secure transmission of data, 
duplicate message suppression, guaranteeing FIFO message delivery and 
transaction management are provided to further illustrate the 
end-to-end argument.

While the main guideline is to put more function into the end system, 
it is important to correctly identify the "ends" of the system, which 
requires subtlety of application requirement analysis. An example of 
voice communication of different requirements is given to illustrate 
ends identifying and that the end-to-end arguments should be used as a 
guideline carefully rather than as an absolute rule.

In the end, the authors discuss the history and application background 
of end-to-end arguments.