From: Scott Schremmer (scotths@cs.washington.edu)
Date: Wed Oct 13 2004 - 09:29:10 PDT
This paper makes the argument that as much functionality as
possible should be placed at the end points of the network rather than in
the lower levels. The argument is made that the lower level functionality
might not be useful to all services (thus in some cases the extra
computation could be wasted), it could be more costly to implement this at
the lower level and such functionality might have to implement at higher
levels anyhow.
The paper uses an example of a file transfer system to show that
despite error checking in the communications portion of the system, error
checking must also occur at the end points due to other potential errors in
the system thus making the error checking in the data communication system
redundant. The paper presents a fairly convincing argument using
several examples and does concede the point that there are instance in
which it would be better to implement such things at a lower level.
However, the paper makes little effort to quantify that which it is
discussing. Exactly how much extra computation time would be needed to
implement error checking in the data communication system? Would that
extra computation time save any of the computation time at the higher
level? Despite the fact that the error checking would have to still be
present perhaps it could be less complicated in some way because it has
less error to deal with. The paper could be improved with an attempt to
quantify some of this.
This paper probably played a role in the distribution of functionality
amongst the various network layers. Thus, it is probably useful in
understanding how we got to where we are today and how this could be
improved upon as our constraints change from what they were.
This archive was generated by hypermail 2.1.6 : Wed Oct 13 2004 - 09:29:10 PDT