From: Chandrika Jayant (cjayant@cs.washington.edu)
Date: Wed Oct 06 2004 - 01:05:03 PDT
"A Protocol for Packet Network Intercommunication"
Written by Vinton G. Cerf and Robert E. Kahn
Reviewed by Chandrika Jayant
This seminal paper by Cerf and Kahn explores the interconnection of
packet switching networks by describing a strong and flexible protocol.
The structure of TCP/IP and its implementation are both covered, along
with potential and inevitable problems and suggestions to their
solutions.
There are many great things about this paper. The authors do a wonderful
job breaking down processes, networks, and gateways into separate
entities with separate responsibilities, which aids the reader when
sorting out all of the technical details presented. Cerf and Kahn
clearly explain where individual networks can differ and address each of
those differences in the following pages. The key point of the protocol
being simple to account for flexibility is conveyed beautifully by
separating what must be consistent across networks (i.e. addressing) and
what tuning can be handled by the gateways while routing (i.e. packet
size changes). The paper also addresses potential problems quite
honestly and logically, such as retransmission in times of failure,
packet size inconsistencies, sequence problems, and duplicate detection.
Even classic papers have their problems as well. Though
admitted in the conclusion, it seems almost a cop-out that the protocol
described in such detail has not been experimented with and that there
are no tangible results to look at. This would round off the paper and
make it seem more useful. Accountability is barely touched upon and I
wonder if the authors didn't consider it important at the time or didn't
see a good way to deal with the problem. I was also struck by the fact
that the authors didn't have the foresight to allow for larger TCP
addressing schemes, but then again I do not know firsthand the state of
technology in 1974 so I cannot put myself in their shoes. The gargantuan
internet is so prevalent today (thanks in part to this paper) that I
cannot imagine a world without it.
To improve this paper, at least if the authors didn't do
some concrete experimentation that could yield results, I would like to
see some concrete suggestions as to how these experiments would run and
what the expected results would be. I would also like more discussion on
bounds of how long the sender must wait for acknowledgement of packet
reception. In general, flow control seemed a weaker part of the paper
that could be beefed up with more potential problems and solutions.
It is always important to realize how something we use in
everyday life is structured, not only for the base understanding of what
we are using, but also to help engineers and scientists build on top of
and add to the protocol architecture. It also helps people diagnose
problems more easily if they know the packet network structure in some
detail. Not many papers stay completely relevant 30 years after their
birth, but the fact that this one has speaks volumes.
In terms of future work, Cerf and Kahn suggest producing a
protocol specification. This would encourage experiments with the packet
switching networks to help tweak more parameters and would produce the
tangible results needed.
This archive was generated by hypermail 2.1.6 : Wed Oct 06 2004 - 01:05:17 PDT