> Hmmm ... since we're dealing with probabilities here, it would > almost make sense to say "on a media with packet loss probability > P_loss, send it N times where P_loss ^ N >= P_acceptable". But > then we're stuck trying to make a decision about P_acceptable, > and in any case P_loss is unlikely to be trivially estimable > (and unlikely to be truly independant ...) >
Link layers typically have a base level expected loss rate (For DSL it is 10^-9). Then there's losses due to equipment faults, they would be hard to estimate by a node just coming on the link. P_acceptable could be estimated just as retransmit timer values are, and a base line provided. The actual value would then be configurable depending on link type. But maybe this is too much detail... jak -------------------------------------------------------------------- IETF IPv6 working group mailing list [EMAIL PROTECTED] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------