From: Danny Wyatt (danny@cs.washington.edu)
Date: Tue Oct 05 2004 - 17:30:26 PDT
A Protocol for Packet Network Intercommunications
Cerf & Kahn
This paper presents the original proposal for IP and TCP. (Initially
described as a single protocol, though UDP is parenthetically
foreshadowed on page 640.) It briefly outlines the need to keep the
protocol simple so that it can incorporate any future networks, the
scheme for Internet work addresses (one byte shorter than the ones with
which I'm familiar) and then, more thoroughly, describes TCP (where the
'P' is for program).
As far as strengths go, I found the use of byte offsets as sequence
numbers a particularly clever and elegant way to deal with
fragmentation. Though I'm not sure why it's necessary to preserve the
initial segment boundaries with flags. Couldn't a fragmenting gateway
recalculate it's own checksums? Since acknowledgments are sent using
byte offsets, the receiving host doesn't need to know how the sending
host originally segmented a message.
The largest unproven assumption in the paper is that the protocol will
work at all. Where are the numbers and graphs? How did this work in
simulation? Until someone actually implements this protocol, I'm going
to remain skeptical of its ability to perform well. All joking aside,
they do mention all of this in the conclusion---even the fact that the
protocol is not fully specified yet, so it couldn't be implemented for
experiments. What they do have is obviously well thought out (and it
makes me regret that I missed the days of the IEEE Journal of
Unempirical Speculations).
The paper could be improved by providing more justifications for certain
decisions (like why gateways shouldn't recalculate checksums), and it's
relevance today is obviously beyond anything anyone could have imagined
at the time of its publication.
This archive was generated by hypermail 2.1.6 : Tue Oct 05 2004 - 17:30:31 PDT