Review of Cerf and Kahn paper

From: Kevin Wampler (wampler@cs.washington.edu)
Date: Wed Oct 06 2004 - 03:36:19 PDT

  • Next message: Pravin Bhat: "Review2-PravinBhat"

    In this paper Cerf and Kahn sketch out the ideas for the transmission
    protocol TCP. They describe both the technical details of this protocol
    and the motivations for its features.

    The protocol described shows great deal of though in its design. many
    properties that a protocol might have are considered, and are accepted or
    rejected on well thought-out bases. Early on the need of a standardized
    protocol (as opposed to translations between different protocols at
    gateways), but some flexibility is built into TCP to accommodate different
    networks. Similarly a protocol requiring packets to be of uniform size
    between networks is rejected, but the use of a sequencing number based on
    bytes rather than packet count easily allows the message to be
    reconstructed if a packet is fragmented. Perhaps most elegant is TCP's
    use of the sliding window and acknowledgment schemes to support segment
    sequencing, dropped packets, and flow regulation.

    As far as the limitations of the described protocol are concerned, there
    are the issues with the insufficiency of a 32-bit sequence number and a
    16-bit window size which are discussed in the book. It is also clear that
    TCP is intended only for applications where a reliable byte stream is
    needed. I would imagine that across noisy networks the overhead and
    delays incurred for retransmission of lost packets and acknowledgments may
    be significant, particularly in applications such as voice over IP where
    data loss during transmission is not as critical an issue as timely
    delivery.

    This paper clearly remains relevant, as TCP is still a widely used
    protocol. Since networks continue to evolve, future work will no doubt be
    of many different varieties. The textbook mentions some simple extensions
    which solve some of the problems arising from the limited size of some of
    the fields in the protocol. Given that multimedia over the Internet seems
    to be an increasingly important capability, I might also imagine work
    being done on more advanced flow regulation control, such as the ability
    to specify the allowed data loss that might be tolerated for the sake of
    improved speed of transmission.


  • Next message: Pravin Bhat: "Review2-PravinBhat"

    This archive was generated by hypermail 2.1.6 : Wed Oct 06 2004 - 03:36:20 PDT