Reacting to previous messages

["David R. Cheriton" <cheriton@DSG.Stanford.EDU>]

Folks,
A new WG is proposed precisely because there appears to be sufficient
interest and motivation for exploring new models of multicast.
I think it is quite fair for the PIM crew to attempt to extend PIM-SM
in whatever way they care to, including to address concerns that come
up in this effort.  If this WG cant come up with a compelling model,
protocol, etc. with PIM as competition, Darwin will take care of it.
So, agreeing with cdiot, I'd like to see a separate effort and protocol.
I also dont see a problem with co-existence or any reason this should
slow the deployment of current multicast.

In my thinking as an ISP (Irreverent Stanford Professor :), multicast
is far less compelling for many applications than believed, and
difficult to manage and package as a product, limiting its deployment.
For example, web caches allow efficient 1-to-N distribution in many
cases with the additional benefit of allowing on-demand access.
Some web caches do/will support caching & splitting of even video
streams, reducing the need for multicast at the IP layer.
Thus, I think we should take the most compelling app., define the
most basic model for that app, and provide adequate facilities to
manage and package it as a service and sink or swim based on whether
that produces a coming solution or not. As I stated before, I think
Internet TV is the most compelling app.

In this vein, Cheng-Yin Lee'w work proposal mentions targetting
"these type of applications:
* Internet wide information distribution (eg 'webcast', content
subscription/distribution)
* Internet wide interactive information exchange (eg
conferencing/discussion, games)."

I dont see games as compelling because:
i) it is a small-scale application - a small number of members,
   and can be done with a unicast simulation of multicast through
   the commonly used game servers.
ii) it is a small market, has been for years, with lots of failed
  companies to prove it.  Until more of the world turns into
  affluent 13-year-old teenage boys or their equivalent, I dont expect
  much compelling action here, at least relative to TV/video.
I also dont see generic webcast, discussion, etc as compelling either.
This is all going on now on top of unicast, and unicast data rates
will go up lots before we reach any of the milestones in the plan,
making them even less compelling.

In contrast, Internet Ted Turner sends a 6 Mbps MPEG stream to
100 million subscribers in real-time gives you a 600 terabit
per second delivery problem.  Ted will pay a lot for this service,
and clearly multicast would be an enormous win.

This WG needs to get hard-nosed about application requirements.
A non-trivial part of the problem with the multicast 
evolution to date is lack of recognition of what are the
really compelling applications.  In that vein, I tried earlier to 
prompt some discussion of what applications compelled the need for
bi-direction shared trees.  I think they are simply not worth doing,
nothing came up to suggest the otherwise, and yet they seem back
on the table as assumed useful.

Personaly, if there is no more agreement in this direction from 
others, I'd feel happier pursuing a separate WG for Internet TV and
doing multicast in that context.  I think the result would still
serve all the "real" multicast applications, but would be better
tuned to scale and package as a very-large scale service.

So, how about the first step in "work" as: agree on compelling
multicast applications, i.e. compelling for a new multicast service.

David Cheriton