Jon Crowcroft wrote:

Gonna say, ironically, one early use of multicast was a proposal to use SRM
instead of a mesh of tcp connections for iBGP...so some people do think
about scaling control plane traffic in the presence of congestion, some
times:-)
That is a terrible approach, because, even within a link, TCP
mesh is the way to go against congestion.

That is, within each link, routers should not rely on link
multicast to exchange multicast control messages and should
have TCP mesh between them over which the messages can be
exchanged reliably even if there is congestion, which is
what I did in 2001 with SRSVP (Simple RSVP, a stable and
hierarchical unicast/multicast QoS routing protocol
without crank back).

                                        Masataka Ohta

Reply via email to