From: Lillie Kittredge (kittredl@u.washington.edu)
Date: Tue Oct 05 2004 - 22:18:26 PDT
A Protocol for Packet Network Intercommunication
This paper is basically TCP/IP's embarrassing baby pictures that its mom
shows us when we come over for coffee. (Continuing the stupid analogy,
this snapshot is of TCP/IP when they were still conjoined, before the
surgery to separate them.)
This represents a much earlier phase in the development of the Internet
than we saw in the last paper. Here we are just starting to consider how
existing networks might be connected, and seeing the tribulations
associated with this attempt. Whereas current link-layer protocols are
designed with the knowledge that they will have IP over them, here we see
the difficulty of creating IP in the first place. The fact that the
existing networks they were combining were so different is probably a
blessing to us today - it means that IP is flexible enough to deal with a
variety of link layer protocols.
I rather enjoyed the discussion in the "process level communication"
section about where to divide byte streams coming from a source. The
fact that they even consider the option of making packets of arbitrary
chunks of the stream shows how early this technology was. It seems
obvious to us now that of course data needs to be parceled by application
and in a way that will make it easy to reconstruct should it come out of
order.
My biggest complaint about this paper is the apostrophe-plurals. Man,
that boils my blood. Also, they misuse the subjunctive.
On the substantive front, however, the major shortcoming of Cerf and Kahn
is that they were trying to bite off more than they could chew. Though
this defines all sorts of stuff which we later see in TCP, such as the
sequence numbers, it's also providing the same service IP does now,
dealing with getting packets from one network to another. On the other
hand, at the time they were fairly reasonable in their decision not to
separate the protocols (or, more likely, in the fact that separating them
hadn't occurred to them).
This is relevant today by the way it provides perspective on how our dear
widdle TCP/IP has grown up into a great big protocol. And it's fun to see
things like the bit about "8 bits allows up to 256 distinct networks.
This size seems sufficient for the forseeable future," which makes me
giggle.
This archive was generated by hypermail 2.1.6 : Tue Oct 05 2004 - 22:18:26 PDT