All things related to multicast

[Havard Eidnes <>]

Hi,

while I can certainly sympathise with the goal of simplifying
multicast by abolishing ASM, RPs, MSDP and all the associated
complextity, I see practical problems in doing so in the short term.

In addition to getting SSM and IGMPv3 "everywhere", there's the
problem of IGMPv2 and IGMPv3 interaction which needs to be dealt with.
If I've understood correctly (section 7 of RFC 3376, IGMPv3), IGMPv3
and hence SSM has what I would call a massive deployment bug for
shared networks: if a single node on a subnet speaks IGMPv2, all the
IGMPv3 speakers on that subnet are by protocol design mandated to
downgrade their operation to IGMPv2, which directly translates into
"No SSM for you!" (or any of your neighbors on that subnet for that
matter).  Thus, hooking up a single IGMPv2-speaker to a SSM multicast-
enabled subnet will act as a DoS on all the SSM-users on that subnet.
I seem to recall seeing various crutches and kludges possible to use
to work around this, but if the goal was to reduce kludges and
complexity and improve common standards and interoperability, that's
exactly the wrong recipe.

Of course, the IGMPv3 spec can at least in theory be updated so that
this isn't the specified behaviour anymore, and IGMPv2 (and v1) can be
relegated to historical status along with ASM, RPs, MSDPs and all the
associated protocol machinery, but we would then need to go through a
full re-implementation and subsequent deployment cycle for all the
IGMPv3 speakers (and IGMP snoopers?) for this to be effective, and
this is going to take ... quite a while.  Or is this not a viable
path, e.g. due to protocol reasons (IGMPv2 hosts becoming confused by
IGMPv3 traffic?)?

Over in IPv6-land things don't seem to be much better, at least not in
principle, there we have both MLDv1 and MLDv2 and MLDv2 also has this
"you need to speak the least common denominator" deployment bug which
IGMPv3 suffers from.  In that context the blessing that we don't have
MSDP for IPv6 is meagre consolation.  Or does ~every IPv6
implementation also do MLDv2?

Anyone wanting to carry that torch?  Or is Internet Mcast a hopelessly
lost cause anyway, independent of IPv4 or IPv6?  Or is the answer
different depending on protocol version?

At least "plain IPv6 unicast", doesn't suffer from any corresponding
deployment bug, if that's anything to learn from.  Imagine where we'd
be with IPv6 unicast deployment today if a single IPv4 host could
through its mere existence veto IPv6 use on that subnet...

Regards,

- Håvard