From: Craig M Prince (cmprince@cs.washington.edu)
Date: Mon Oct 11 2004 - 03:29:11 PDT
Reading Review 10-13-2004
-------------------------
Craig Prince
The paper "End-To-End Arguments in System Design" was very much a paper
discussing general good design, rather than a networks paper; however, the
focus of many of the examples were around the design of networks and
network protocols. The main thrust of the paper is on how if you design a
feature(s) into the lower-levels of a system you are bound to have some
application that does not require or requires a modified version of this
feature and ends up re-implementing it. Another basic idea of the
end-to-end argument is that placing features at a lower level may provide
some useful service or assurance for that lower-level, but does not
provide any assurance outside of that lower-level, instead many
applications need a feature or assurance from "end-to-end" and so the
feature is reimplemented at each endpoint.
The paper also brings up the good point that many times features are
placed in the lower-levels of a system for performance reasons; however,
there are often better performance gains to be had by placing the feature
at a higher level (where additional information can be leveraged).
While this paper mentions the pitfalls of placing features/functions at
too low a level, I felt it could have done more to provide intuition into
when to place features at a high or low level. The authors gave many
examples of different applications having different requirements and also
how applications had to reimplement features even though part of the
system already provided the feature; however, I was left with little
indication of what actually should be included at the low-level.
This paper gives interesting guidance into good design for network
protocols and gives some motivation behind the internet protocol layerings
since layering allows features to be separated from low-level protocols.
This is something to consider when designing future protocols.
This archive was generated by hypermail 2.1.6 : Mon Oct 11 2004 - 03:29:11 PDT