[e2e] random early time-to-die

John Kristoff jtk at depaul.edu
Thu Jun 14 08:31:21 PDT 2001

Jon Crowcroft wrote:
> here's an odd idea

Yep.  :-)

> what if we decrement TTL by an amount proportional to the marking
> probability (instead of marking or dropping in RED)?

This is *very* random since you don't know how many hops have been
traversed up to the point of marking or how many hops are left to go. 
You also don't know what the initial TTL might be.  Shouldn't the queues
be dealt with at the point where marking happens?  How would this scheme
help routers/networks more effectively than something like RED?  Instead
of randomly dropping at the point of marking, the dropping is delayed
further into the network (where you have no view of what the conditions
are).  I'm not sure that is a good thing.


More information about the end2end-interest mailing list