From: Karthik Gopalratnam (karthikg@cs.washington.edu)
Date: Tue Oct 05 2004 - 22:49:31 PDT
Review 2 - Cerf and Kahn - A protocol for Packet Network Intercommunication
This paper is the first description of the TCP-IP protocol. It describes
the various engineering decisions that went into the creation of TCPIP, and
the problems that it attempts to solve.
This paper is motivated by the early internet - a small number of
disparate networks each designed and administered in vastly different ways,
but with resources which, if shared could result in mutual benefit. The
overarching goal here is to build a protocol which would enable these
disparate networks to talk to each other in some common "language", while
still preserving each network's inherent autonomy. This requirement resulted
in the design of the gateway, which would form the "core" of the
internetwork, and perform the work of actually forwarding packets to the
correct destination network.
The major contribution of this paper to the Internet as we know it today
is TCP in an early form. The authors introduce TCP basically in almost the
same form as we know it now - handshake, sequencing and acknowledgements for
reliable data transfer. Of course, it is understandable that the deamnds of
the internet at the time of this paper did not include anyting about
congestion, so none of the congestion handling logic appears in the design
of TCP. However, even though the authors do not state it as an explicit
goal, and probably did not consciously intend it either, a fundamental
conribution of this design is that TCP essentially puts the onus of managing
data flow on the end hosts, thus leaving the core to perform just the
routing functions. This is a fundamental strength of network designs today.
It is surprising, but I suppose understandable to some extent, that the
authors lacked the foresight to expect that the internet might consist of
more than 256 networks some day!! (Of course, *we* have the benefit of 20-20
hindsight!) Perhaps if they had anticipated the need for more
interconnections, they would have seen the benefits of making TCP separate
from IP, whereas as per their design, TCP-IP is really a monolithic entity.
That being said, many of their design decisions have stood the test of time
and scale, and this is clearly one of the classic papers of networking.
This archive was generated by hypermail 2.1.6 : Tue Oct 05 2004 - 22:49:32 PDT