From: Craig M Prince (cmprince@cs.washington.edu)
Date: Wed Oct 06 2004 - 01:37:37 PDT
Reading Review 10-06-2004
-------------------------
Craig Prince
The Cerf and Kahn paper titled "A Protocol for Packet Network
Intercommunications" is a classic networking paper that describes the
initial proposal for the TCP internet protocol. This paper outlines the
basic requirements required by the TCP protocol -- namely to be a simple,
versatile protocol that would be robust to lost packets and work across a
variety of types of networks. The paper not only outlined the structure of
the protocol packets, but also discussed the motivation behind the
protocol as well as the procedure for setting up connections and
communicating between processes.
One strength of this paper is the fact that it indirectly introduces the
notion of protocol layering. To show how their protocol could be used
across existing networks, the authors embed their protocol header within
the data of the individual local network packets. This layering allows the
unification of different networks without requiring those networks to
change themselves. Another strength of this paper is that it presents
alternative design choices and effectively argues as to which choice is
best. For example, the paper discusses two cases regarding how to handle
packets from/to different processes on the same machine, describing the
pros and cons of each.
I had a few concerns while reading this paper. The first, is that while a
coherent protocol is presented there is no alaysis of the correctness of
the protocol; for example, are there unrecoverable states?...when will the
protocol break? Another concern that I had with this paper is that it
makes the assumption that all communication will require delivery
quarantees. While this is true for many applications, this is not always
the case. By including this directly in the internetwork protocol, this
overhead is always incurred. This concern is one of the reasons that a
separation between TCP and IP was finally made. This work could have been
improved by giving a more rigorous handling of the proposed protocol to
prove that the desired properties held for it (Will fragmentation and
reassembly work in all cases? Can the window and sequence number become
unaligned? Etc.)
This paper is relevant today in that it gives a good overview of the
design ideas behind the TCP protocol. It is interesting that the TCP
protocol still used today would have such a close resemblance to this
initial proposal made in 1974. The great care that was taken to ensure
that the protocol was as unrestrictive as possible has led to its
longevity and serves as a guide for future protocol designs. This paper
doesn't touch on features such as security or quality of service, which
suggests that there remains work to find protocols that include these
features as well.
This archive was generated by hypermail 2.1.6 : Wed Oct 06 2004 - 01:37:37 PDT