Re: Request for review of draft-wu-rgmp-01.txt

[WG participant hat firmly planted on head]

When I first heard of RGMP, I wondered why IEEE 802.1p's GMRP
did not suffice.  RGMP does use IP multicast addresses, instead
of ethernet multicast addresses, but there may be little benefit
from that because routers aren't allowed to send LEAVE messages
for groups that overlap with other groups that the router does
want traffic from.

GMRP also handles multiple switches gracefully, etc, something
that's not completely handled in RGMP.

I'd like to see the spec discuss why GMRP is not sufficient and
why RGMP is better.


On the topic of the spec itself, I think that it should admit that
it's PIM-specific, instead of saying that it is general and then
saying that you could maybe use non-PIM mechanisms if you felt like
it.  It doesn't even suggest using multicast router discovery for
discovering multicast routers, instead saying to use PIM protocol
messages.


I think it's a shame that RGMP continues the 224.0.0/24 problem, but
since we're not allowed to suggest protocol changes, such is life.
Hardwiring 224.0.1.39 and .40 is also unfortunate; that's a Cisco
protocol that a given network might not be running.


RGMP happened to pick IGMP type numbers out of the space that's
reserved for experimental usage.  I guess that's OK, except that
collisions might occur and presumably fairly unpredictable behavior
will result.  However, again, since we're not allowed to do protocol
changes, there's nothing for us to do here except live with it.


  Bill