From: Ioannis Giotis (giotis@cs.washington.edu)
Date: Tue Oct 12 2004 - 23:30:56 PDT
A major question when building layered protocols is at which layer to
implement several functions such as error detection, encryption and other
services. The authors focus on several functions which according to their
claims should be implemented on upper layers.
The main idea is that by incorporating these functions in the lower levels
there is a "gap" between the function and the application actually using it
which results in inefficiency. The authors go through various examples to
illustrate that it is better for the end points to actually handle most of
the functions and even demonstrate some real-life applications. Their main
argument is that is safer and easier to handle these functions through the
application.
The authors have a vague understanding of layering (they do mention the word
in the very last paragraph), however it seems that they're dealing with a
much more simplified network than today's examples. They propose to
implement functions at the application layer without considering all the
intermediate ones. Although it is true that application layer functions
would make the connections more flexible and easier to debug and maintain
they fail to mention the added complexity involved.
It is true that if the Internet had followed such an approach a lot of
today's network uses would be easier to handle, however no one can say if
we'd ever reach this point and the difficulties developing complex network
applications wouldn't be greater. To give the paper a point, it is true
that today some functions are implemented in the application layer however
the motivation was totally different and had to do with backwards
compatibility.
Overall, one should be careful when making design decisions as to where to
implement several functions. It seems that there are always advantages and
disadvantages in pushing a function higher of lower in a layer hierarchy.
The final decision should not be a fixed design principle but rather a
careful balance of the desired effect on the network.
This archive was generated by hypermail 2.1.6 : Tue Oct 12 2004 - 23:30:56 PDT