HN-SPF Review

From: Karthik Gopalratnam (karthikg@cs.washington.edu)
Date: Sun Oct 31 2004 - 23:40:36 PST

  • Next message: Tyler Robison: "Review of "A Revisd ARPANET Routing Metric""

    Review: A Revised ARPANET routing metric

       This paper presents a new routing metric to aid routing decisions - the
    HN-SPF metric. This is intended to fix the instability problems with the
    earlier D-SPF metric. The authors note that under heavy loads, the D-SPF
    metric, which is based on estimating instantaneous delay from neighboring
    nodes, can lead to oscillations in routes, and also causes links to be
    underutilized as routes tend to all shy away from heaviliy loaded links at
    the same time. The authors' solution to this is the Hop-Normalized SPF
    metric, which is based on using a virtual hop as a cost unit for a link, and
    then normalizing the cost across various links. This enables the authors to
    prove in a nice mathematically rigorous way using equilibrium points, that
    the fluctuations of the HN-SPF metric are minimal under heavy load. On the
    other hand, this metric performs just as well as the original delay metric
    under light loads. Another feature of this solution is that it is orthogonal
    to the underlying method of finding routes, and merely adds the (crucial)
    delay metric component.

      Overall, the authors have done a great job in indenftifying a problem with
    the existing ARPANET implementation and proposing an elegant solution.
    However, the choice of the parameters for the 8 lines in their proposed
    solution, is somewhat of a black art. Of course, it can be argued that this
    gives designers more leeway in making things work optimally, but it seems
    that these parameters have to be tuned really well for the true strengths of
    this solution to be exploited.


  • Next message: Tyler Robison: "Review of "A Revisd ARPANET Routing Metric""

    This archive was generated by hypermail 2.1.6 : Sun Oct 31 2004 - 23:40:38 PST