From: Jenny Liu (jen@cs.washington.edu)
Date: Mon Nov 01 2004 - 03:41:16 PST
"The Revised ARPANET Routing Metric" presents a revised metric for
packet switching nodes to determine relative link cost (and thus best
route) for packets.
The revised metric fixes some of the major problems of the old (delay)
metric. In particular, utilzation of links in a network employing the
new metric is able to oscillate with greatly reduced amplitude (there
was no cap at all on the amplitude of utilization with the delay
metric), resulting in greater overall utilization of network bandwidth.
In addition, packet switching nodes employing the new metric create
fewer frivilous routing updates, and thus create less routing overhead.
The revised metric fixes some of the major problems of the delay metric,
but the improvement seems too little, not drastic enough. Further, it's
not clear that the system will behave well when the underlying
assumption (about the flow of traffic on and off a considered link
during the process of determining its equilibrium not significantly
affecting the costs of other network links) made in the analysis of the
system may not hold (when link utilization is high).
The revised metric was a viable fix to a problem at hand, but perhaps
does not push the envelope hard enough.
This archive was generated by hypermail 2.1.6 : Mon Nov 01 2004 - 03:41:09 PST