Review #9: The revised ARPANET routing metric

From: Rosalia Tungaraza (rltungar@u.washington.edu)
Date: Sun Oct 31 2004 - 21:27:40 PST

  • Next message: Kevin Wampler: "Review of A Revised ARPANET Routing Metric"

    This paper describes the Shortest Path First (SPF) algorithm and its two
    variants; the Delay-SPF and the Hop Normalized SPF (HN-SPF). Both
    algorithms are used in networks for routing purposes in which they try to
    find the best (cheapest) and thus shortest path for a packet to traverse
    the
    network from one node to another. Their major difference is in how they
    define the best path. The D-SPF defines the best path to be any path that
    will minimize the delay of the packet from one node to another, while the
    HN-SPF defines the best path in a similar way, but also takes into account
    issues such as link utilization, line-type, and line-speed.

    The HN-SPF algorithm was meant to be an improvement to the D-SPF
    algorithm. Hence, a means to reduce the shortcomings of the latter
    algorithm, which included the fact that under heavy traffic much network
    resources was wasted due to their inefficient allocation, and the
    continuous possibility of congestion because some subnets were
    over-utilized.

    One of the successes of this paper and thus the HN-SPF algorithm is that
    it ended up being an improvement to the D-SPF algorithm. For instance,
    when the performance of both types of algorithm was tested by taking into
    account their behavior at system equilibrium states, the following was
    seen: the D-SPF tended to be either oversubscribed when the system (link)
    has heavy traffic and idle when the system had light traffic at
    equilibrium. On the other hand, the HN-SPF tended to be relatively stable
    (oscillates around it equilibrium) in both situations at equilibrium.

    As stated in the paper, one of the limitations of the HN-SPF algorithm is
    that it is mostly effective when network traffic contains several small
    node to node flow. Thus, I think one possible future work is to either
    improve this algorithm further to make it equally effective in other
    scenarios as well or to come up with a completely new algorithm that can
    do so i.e. a multi-path routing algorithm.


  • Next message: Kevin Wampler: "Review of A Revised ARPANET Routing Metric"

    This archive was generated by hypermail 2.1.6 : Sun Oct 31 2004 - 21:27:52 PST