[e2e] random early time-to-die

Steven Berson berson at ISI.EDU
Thu Jun 14 09:26:11 PDT 2001


A problem with this scheme would seem to be that decrementing the
TTL at the congested router may not quickly enough fix the congestion
at that router, so here is an odd variation:

Rather than proportional to the marking probability, have the TTL
decrement be a random variable.  The mean of the random variable would
be an increasing function with the congestion.  But there would be a
(constant??) variance that would cause more packets to be dropped than
with strictly proportional TTL decrement.

People could then ``pay'' for extra TTL.

P.S.  Note that this will adversely affect traceroute.

On Thu, 14 Jun 2001, Jon Crowcroft wrote:

> here's an odd idea
> what if we decrement TTL by an amount proportional to the marking
> probability (instead of marking or dropping in RED)?
> j.

More information about the end2end-interest mailing list