From: Daniel Lowd (lowd@cs.washington.edu)
Date: Wed Oct 06 2004 - 01:17:11 PDT
This paper describes a packet-based protocol for internetwork
process-to-process communication. It covers issues of interoperability
and reliability through a detailed discussion of gateways and flow
control. The protocol described is sufficiently powerful and flexible
that something like it is still used, 30 years later.
This paper did have a few assumptions and oversights, however. Most
notably, no attention is given to security. Although management of the
diverse networks is assumed to be distributed, hosts and processes are for
the most part trusted to play by the rules. Furthermore, 256 networks
each with 65,000 host/port combos is assumed to be enough -- perhaps that
was generous when the paper was written, but it is certainly no longer the
case as we run low on IPv4 addresses (of which there are 4 billion) and
transition to IPv6.
Even so, this paper remains relevant today as an early embodiment of ideas
that later evolved into our current protocols.
Other thoughts:
I was surprised to see the protocols described in terms of processes
rather than computers. I usually think of networking as connecting
computers together so that they can share information... but of course,
what really matters is specific processes being able to communicate via
that network. Therefore, the focus on processes is perfectly logical, but
it ran counter to my naive biases.
There was also a lot of emphasis on getting different networks to
interoperate -- allowing for splitting packets, adding header and footer
information locally, etc. This was an interesting contrast to how the
internet is mostly used today, in which most everything runs IP. The
"local networks" that need to be internetworked seem to be diminishing as
more and more hosts run IP.
Finally, the discussion of accounting surprised me. How much accounting
do ISPs really do? Does anyone really count packets and charge per
packet? Perhaps this is just something I have been shielded from, since I
have never used internet services that imposed traffic limits.
This archive was generated by hypermail 2.1.6 : Wed Oct 06 2004 - 01:17:12 PDT