The IP over ATM Mailing List Archive by date

[Grenville Armitage <gja@thumper.bellcore.com>]

Jed,

You're not alone. The support for multicast as it currently
exists in the ATM standards is not as advanced as we might
wish it to be. 

	[..]
>>It seems to me that what is needed is some sort of multicast
>>virtual circuit.  With such a circuit, nodes could join locally
>>and have their cells routed (through a table in the ATM switches
>>much like with unicast or point to multipoint VCs) to the appropriate
>>next switches (much like as now happens with IP multicast).
>>Specifically cells would not be routed back to where they came
>>from (e.g. as happens with the MCS approach).

I believe UNI4.0 (and Phase ?? PNNI) will support the notion of
ATM multicast group addresses, and allow endpoints to splice
themselves onto existing pt to mpt VCs. This will certainly
improve the data distribution path when multicasting across
multi-switch networks (wrt ipmc-07). However...

>>  Also, only one
>>such structure would exist in any switch for a multicast group
>>(unlike with the mesh of VCs approach which has one point to
>>multipoint VC for every leaf node transmitter in a multicast group).

...it is not entirely clear that you can easily construct a mpt-mpt
VC service (to avoid the need for pt to mpt VC meshes).

The major difference between IP multicast routing and ATM multipoint
VCs is that IDMR is routing self-contained entities with full
source identification per packet. In ATM the unit being routed/switched
is the cell, containing only path identification (VCI/VPI). This
is a problem when using AAL5 - the reassembly engine at a/each receiver
assumes a single segmentation engine produced a sequential flow
of cells associated with one AAL_SDU. AAL5 could not operate over a mpt-mpt
VC, as each receiver would be trying to reassemble AAL_SDUs from
multiple sources whose cells were intermingled within the network.
AAL3/4 incorporates a MID field that could be assigned uniquely
to each sender on a mpt-mpt VC (up to 1023), but this AAL has been
pushed aside by history (at least for the time being).

Given the desire to use AAL5, I think mpt-mpt will continue
to be constructed from overlain pt-mpt VCs even after leaf
initiated join signalling is deployed.

Perhaps Joel would have some pointers to ATM Forum/PNNI related work
on this topic?

cheers,
gja