[e2e] random early time-to-die

Joe Touch touch at ISI.EDU
Thu Jun 14 07:57:49 PDT 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)?

Dropping has two immediate effects - reducing the queue length
at the router where congestion occurs, and losing the
packet (hopefully causing the source to backoff).

TTL reduction inhibits the former in all cases,
and the latter for packets already 'near' their destination.
(see caveat in the last paragraph, though).

About the latter, why should that matter? Whether a packet
has 1 or 100 further hops, it's clearly contributing to the
queue congestion at _this_ router. The source needs to be
encouraged to backoff.

One final point - this doesn't correlate to the likelihood
of a packet being delivered before being dropped, excepting
only 'expanding ring search' packets. Sources don't tune their 
TTL to "barely reach" the destination.


More information about the end2end-interest mailing list