[e2e] random early time-to-die

Ted Faber faber at ISI.EDU
Thu Jun 14 09:28:23 PDT 2001


On Thu, Jun 14, 2001 at 03:25:20PM +0100, Lloyd Wood wrote:
> On Thu, 14 Jun 2001, Jon Crowcroft wrote:
> 
> > what if we decrement TTL by an amount proportional to the marking
> > probability (instead of marking or dropping in RED)?
> 
> endhosts immediately start setting TTL to random values above 200.

Actually I think endhosts would set TTLs deterministically to 255.  I
would.

In addition to all the problems Lloyd mentions, I don't see any
advantage to doing this as opposed to doing RED.  They seem to be
equivalent work.  I wouldn't be surprised if you could define a
near-equivalent system by fooling with TTLs, but I don't see any
advantage, though (and a big disadvantage coming in the next
paragraph).  I'm happy to hear if I'm wrong.

More importantly it needlessly overloads the TTL function (again).  If
you want users to pay for lower marking probability or some other RED
parameter that amounts to congestion-based pricing, add an option.
TTLs are already used for loop prevention and scoping.  Adding
congestion-based pricing is a lot to ask a 1 byte-field to carry.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 230 bytes
Desc: not available
Url : http://www.postel.org/pipermail/end2end-interest/attachments/20010614/1a5ad1dd/attachment.bin


More information about the end2end-interest mailing list