Cell Relay Archive

["Scott Brim" <swb@newbridge.com>]

Bert, for non-realtime traffic you packed many cells in a frame.  Real time
traffic would get one or a few cells in a frame.  This would be based on VC
QoS parameters.  You're wrong about "no QoS at all", by the way, since even
if you used full-size frames the maximum delay over a switched 10Mb Ethernet
segment would be about 4 milliseconds.

Go check out the latest incarnation of the concept, the ATM Forum's
frame-based ATM over Ethernet.  You might also find some mention of it in
the ADSL Forum documents.  Or contact Jim Harford (AdvanceNet, I think).

Also check out frame-based ATM over Sonet, once known as FNNI.  This is a
different animal, don't lump them together, but it's the most interesting
spin yet.  It has allowances for non-realtime traffic, realtime traffic,
OAM/RM cell "playthrough", and even reinsertion of OAM performance
measurement cells at the appropriate place in the cell stream when
interworking between cell-based and frame-based links.

Finally, there's no reason why you couldn't use the concepts on gigabit or
10G Ethernet.  They're point-to-point encodings and framings for variable
length frames.  Piece of cake.

...Scott

<albert.e.manfredi@boeing.com> wrote in message
7k8q7p$rv0$1@nnrp1.deja.com">news:7k8q7p$rv0$1@nnrp1.deja.com...
> In article <7k8lau$q9r$1@panix2.panix.com>,
>   shore@panix.com (Melinda Shore) wrote:
> > In article <7k8h97$ns3$1@nnrp1.deja.com>,
> >  <albert.e.manfredi@boeing.com> wrote:
> > >Cells in frames was a concept championed by Cornell University, among
> > >others. What it was (maybe still is?) was a way of using Ethernet to
> > >carry ATM cells, only for the purpose of testing ATM while using
> cheap
> > >Ethernet hardware.
> >
> > Close.  We were doing it to carry voice.
>
> Huh??
>
> At the time, Scott Brim described cells in frames as I stated above. If
> you intended it to carry voice, I'm truly at a loss. Isn't it
> infinitely easier to carry voice over Ethernet using something much
> more standard, like UDP/IP?
>
> Since filling up an Ethernet frame with ATM cells offers no QoS
> advantages above what a stream of large Ethernet frames would provide
> (i.e. basically no QoS guarantees at all), it would seem to me that if
> your motive was to carry voice, using small UDP/IP packets (e.g. 200
> bytes max) would be a far better choice (assuming you're going across
> routers)?
>
> Perhaps what you mean is that you carried voice in the ATM cells in
> order to conduct your tests? The ultimate goal being the tests, over a
> lightly loaded Ethernet to emulate an ATM medium, rather than the goal
> being "to carry voice over Ethernet"?
>
> --
> Bert
> manfredi@arl.bna.boeing.com
>
>
> Sent via Deja.com http://www.deja.com/
> Share what you know. Learn what you don't.