From: Chuck Reeves (creeves@cs.washington.edu)
Date: Wed Oct 13 2004 - 07:15:30 PDT
The paper "End-To-End Arguments in System Design" was written in 1984 by a
number of researchers for MIT. It defends the application of a principle
called End-To-End Arguments to the design of networked computer systems. The
principle essentially claims that many functions necessary to support
application are not appropriate to be integrated into the lower levels of
the system. It goes on to identify specific examples such as encryption,
duplicate message detection, out-of-order handling, and guaranteed delivery
as functions that are more effectively and necessarily implemented at the
application level. I thought the authors the scenarios were effective in
justifying requiring many of the functions described to be implemented at
the application level. However, I thought there were two aspects of this
proposal that were not given enough consideration:
1. What about application complexity. No consideration is given to what
burden this simplification of the communication system's role has on other
layers.
2. The paper briefly mentions Lampson's SOSP Open System paper from
1979, but I think this approach really is what the author is making a case
for. That being, a system with optional behaviors controlled by the calling
application.
Perhaps not surprisingly this paper, clearly supports Clark's concept of
"fate-sharing" as proposed in his 1973 paper on the design of the DARPA
Internet.
This archive was generated by hypermail 2.1.6 : Wed Oct 13 2004 - 07:15:34 PDT