(no subject)

The Scalable Reliable Multicast protocol, which is described in this
paper, is a protocol for the reliable and efficient multicast
delivery of data. It adds reliability to multicast sessions in a
manner somewhat akin to TCP adding reliability to IP, albeit with
different mechanisms.

The central element of the Scalable Reliable Multicast protocol is
the use of static application data units (ADUs). Whereas TCP packet
numbers are unique across streams, ADUs remain constant, so the same
ADU from the same source will contain the same information, allowing
the sharing of data. When a receiver misses an ADU, it will
multicast a request to the group so that anyone on the group can
reply with the packet.

My concern with this protocol concerns what happens when you have
very large multicast groups with significant overall packet loss.
Assuming the packet loss over a large group is relatively non-uniform
(i.e. the packets lost at receiver A may have nothing in common with
the packets lost at receiver B), then packets will be retransmitted
to many receivers who do not need them. A scenario would be
conceivable where every packet is lost by at least one receiver at
least once and as such all packets must be retransmitted. Perhaps it
would be better instead to not multicast retransmissions.
Received on Tue Oct 31 2006 - 08:01:23 EST