Re: Simple Multicast - building a case for a BOF or WG

[Tony Ballardie <ballardie@dial.pipex.com>]

Having just caught up with all the messages on this subject, one thing
that comes across is there still seems to be an underlying desire to
create/define a magic protocol  that fits all applications and scenarios.
If there's one thing we've learnt over the years its that there will (in all
likelihood) never be such a protocol.

The multicast "service model" is also mentioned a lot; the only place
I've ever seen it defined is in Steve D's thesis. Reading between the
lines of the msgs there seem to be different interpretations of what the
multicast service model is... as a starting point I think we need to be in
sync with what the definition is, then decide if/how/under what
circumstances that model needs extending/changing. The service
model is then the basis for further development.

I agree with the comments that the lack of apps that truly require IP
multicast hasn't helped its development. That's changing now to some
extent with Internet TV and a few others, but they're all single source
apps.

For the past six years we've been trying to develop IP multicast protocols
in the hope of converging on a single protocol (in IDMR at least),
assuming one service model, and without even having a well defined set
of target applications... we've developed all these protocols in the hope
that applications will turn up one day that the protocols will be ideally
suited for and which fit the service model. Dave C and Hugh H seem to
have been the only folks who have developed a protocol with a particular
app and operating scenario in mind. That would seem the logical way of
doing it (with hindsight of course :-)

I don't think it would be a bad idea to categorize all potential IP
multicast
apps according to their characteristics and requirements (Bernard
Adoba's Multicast Applications draft has already done this), then map each
to one/more protocols that can satisfy them. There may be "killer apps"
which
turn up for which no protocol is ideally suited, so we'd need to mould an
existing ptcl or develop something new. All this, of course, points to
multiprotocol support in routers, and a means for allowing the apps to
communicate, say, ptcl-type, when joining (by whatever means).

just my thoughts.

Tony