Cell Relay Archive

[manfredi@arl.bna.boeing.com]

In article <5cr8vo$ld0@fang.dsto.defence.gov.au>,
  tim@triton.dsto.gov.au wrote:
> 
> If several recipients of a multicast group are listening to a broadcast 
channel,
> then how can ATM signalling arrange to use the channel to implement the
> multicasting, rather than implementing it in the switch transmitting onto the
> channel?  Most ATM signalling seems to assume that the channel is NBMA
> (Non-Broadcast Multiple-Access).

Yes, ATM is by nature a circuit-switched non-broadcast medium. That's what
allows ATM to make quality of service (QoS) promises, where packet-switched
broadcast schemes like Ethernet and FDDI have a much harder time of it. And
yes, there are tradeoffs.

> For example, we run ATM over satellites for multicast applications such as 
> videoconferencing.  Even though multiple receivers may lie in the footprint
> of a satellite (i.e. broadcast channel), our current switches don't exploit
> this, and instead create a copy of each multicast cell for each recipient,
> which obviously wastes precious bandwidth.  Perhaps, should we treat the
> broadcast channel as a type of switch, and make it responsible for the
> duplication?

Its' not clear than bandwidth is wasted, is it? I mean, we're actually
talking multicast here. I think you're talking about a one-to-many
multicast. What ATM would do is to create a tree. The root is the source of
the multicast (say the satellite receiver). That root creates a point to
multipoint VC, which is in fact as you describe: cells replicated from one
source to many destinations.

At each destination switch, more pt-mpt VCs might be created, until a path
has been established to all who want to receive that multicast.

So are we wasting bandwidth or do you really mean that we seem to be
requiring more computational power than a broadcast medium like Ethernet?

Note what happens with ATM, though. When each switch sets up the pt-mpt
VC it needs, that switch might already have other VCs set up, each with
some sort of QoS requirement. This brand new multicast will not affect
those other VCs. That's the advantage of circuit switched systems. Yes,
setting up the multicast channels does seem more laborious, however.
Because it is.

I'm not sure that I addressed your question, but in fact the desire for
some sort of QoS guarantees is causing new protocols to be created for
packet-switched systems too. And in effect, what happens is that the
packet-switched system takes on aspects of circuit-switched systems.
I'm referring in particular to the Reservation Protocol (RSVP), being
developed by the IETF.

Bert
manfredi@arl.bna.boeing.com

-------------------==== Posted via Deja News ====-----------------------
      http://www.dejanews.com/     Search, Read, Post to Usenet