routing connectivity

[karn%ka9q.bellcore.com@Sdsc.BITnet (Phil Karn)]

Dan's description of the problems inherent in NET/ROM routing reminded me of
a clever and interesting technique I read about some time back in one of the
DARPA packet radio projects for determining link quality.  Perhaps somebody
will adapt it to our needs.
     
Each station keeps track of the number of packets seen from each of its
neighbors, regardless of the destination addresses in those packets.
Periodically, each station broadcasts a packet containing the total number
of packets it sent on the channel during the interval just ended. Each
station hearing the broadcast can then readily determine how reliable the
path is from that remote station to itself by comparing its own count
against the transmitted value.
     
Conversely, if the statistics packet also contains the station's count of the
number of packets seen from each of its neighbors, every other station
hearing the broadcast could determine how well it is being heard by that
station.
     
A routing algorithm could then declare a link "up" or "down" depending on a
comparison between the success percentages and some threshold, or it could
weight several parallel paths based on the probability of success in
each. Combined with a "link state" routing algorithm that distributes the
quality of each link throughout the network, this could be a highly effective
technique.
     
Comments?
     
Phil