Cell Relay Archive

[anouch_atm@yahoo.com (anouch)]

"Albert Manfredi" <albert.e.manfredi@boeing.com> wrote in message news:<GMI93u.2wB@news.boeing.com>...
> "anouch" <anouch_atm@yahoo.com> wrote:
> 
> > This will work as long as the de-jitter buffer at the far end is
>  big
> > enough to absorb the jitter and this may be true almost in all
>  cases
> > provided that the number of switches passed through is limited.
> > Because a regular traffic entering a network becomes bursty at the
> > egress and the more switches in the path, the more bursty will be
>  the
> > traffic at the exit without correcting irregularities at each
>  node.
> > I believe the "deterministic" way to bound the end-to-end jitter
>  is to
> > "control" it at each node and not just to "accomodate" with it.
> 
> I think I agree here. In part also because if the supposedly CBR
> traffic becomes very bursty in the core network, through no fault of
> the source device, it might be policed out of existence. I suppose
> this would depend on where the traffic is being policed. No? This
> discussion happened some time ago in the diffserv wg of the IETF.
> 
> Bert
> albert.e.manfredi@boeing.com

Yes, and that's why we have to shape (at least) at the egress in order
for the traffic pattern to become as it was when entering the network.
Here egress can be toward the very end user expecting a regular
traffic as was sent by its far end side. Egress can also be the
ingress of another diffserv domain where the traffic can be policed.
Obviousely there is no sense for a conforming user traffic at the
ingress to become suddenly non-conforming just because the network
itself introduces jitter which is out of user control.