From: Tyler Robison (trobison@cs.washington.edu)
Date: Tue Oct 05 2004 - 23:26:44 PDT
Review of "A Protocol for Packet Network Intercommunication"
This paper proposes the TCP protocol, which is meant to be the
interface from a process to a network for sending data to computers that
lie in other networks. The process hands the data, destination computer
id (consisting of a network number and computer number), and a port, over
to the TCP protocol, and the protocol handles transmitting it through and
across the networks along the way. Beyond the overall concept, the paper
discusses formats for id's and headers, sequencing techniques,
fragmenting/reassembly, a procedure to retransmit unreceived packets, and
various other aspects of the protocol. One key idea is that the core data
is encased in headers as it goes along; headers which differ from network
to network, and which are used to pass the packet along whatever type of
network it is passing through, while the data itself is unchanged. The
paper appears to be describing TCP and IP protocols as a single protocol.
A significant limitation is that it wasn't based on experimental
data (at least not directly; previous experience probably went into its
design), and it sounds like additional testing was necessary to really
evaluate and complete the ideas. For example, if it wasn't until later
that the protocol was split into TCP and IP, it could be because the paper
didn't consider all necessary information at this point. Its also fairly
difficult for me to read this and identify which ideas are entirely new to
this paper, and which are simple extensions of the networking concepts of
that time. In addition, while fairly detailed descriptions of much of the
protocol are included, some very important parts are left out (which
admittedly were meant to be filled in in later papers), such as how a
gateway is choosen; it leaves a somewhat incomplete picture. The
improvements I would suggest, things to fix the above, have of course
already been done.
The protocol laid out here clearly shows itself to be the ancestor
of TCP/IP, even if various details are a bit strange: the 'TCP address'
as 8-bit network address and 16-bit identifier, for example. Although I'm
not entirely familiar with modern TCP/IP, the various ideas of adding on
headers, the addressing, ports, notion of a protocol right below the
process which handles sequencing, and a dozen other facts all appear to be
almost identical to those of the present, so it seems that the ideas here
were successful.
This archive was generated by hypermail 2.1.6 : Tue Oct 05 2004 - 23:26:44 PDT