Re: [MBONED] WGLC for <draft-ietf-mboned-ipv4-uni-based-mcast-04.txt>

On Tue, Jan 22, 2008 at 11:53:51PM +0100, Stig Venaas wrote:
> The one unique thing where ASM is needed IMO is service discovery,
> autoconfiguration etc. Cases where you need some multicast address or
> perhaps anycast for bootstrapping... But this isn't something you would
> do on the Internet, just inside a site or an organisation.

Hmm.. for service discovery/autoconfiguration ASM is convenient to the
application if then available in the netork, but not necessarily the
best solution or only solution. anycast or dns or lots of other mechanisms can
be used for service discovery etc. as well.

I think the ultimate issue with ASM is still the security from unwanted
sources. Nobody has clearly answered me yet how it can be within the
security requirements of the IETF to do something like ASM across 
a completely untrustworthy scope like the Internet (without creatng
a mandatory spec dependency against a security layer - and just
saying IPsec is not sufficient, because it would only allow
discarding of unwanted traffic due to missing authentication _AFTER_
that traffic has wasted bandwidth to a potentially large number of
receivers).

I guess we never faced this challenge from the IETF, because rfc1112
was never updated ? Like to include a specification that the ASM service
model also applies to IPv6 ? I guess i digress...

I think that most common ASM applications do have a client/server
structure and they should be using a unicast to server, SSM back from
server transmission scheme to allow for better security (as in: server
can easily control sources).

Obviously, ASM is very convenient to applications, so as soon as
the scope is sufficiently well controllable, there's nothing bad
about using ASM. Those scopes are usually structured such that
one can also more easily move to Bidir. Let's call them enterprises.

> In other cases I think you can do source discovery at the application
> layer and use SSM. Of course there is lots of complexity in doing it
> in the application as well, but I think that is a better approach.

Somehow i remember a time, let's call it the 80th and early 90th where
most of the protocols and networked applications running on the
Internet where based on reuse of research developed libraries and application
code, freely available. I wonder where that model went for the simple
object of an application layer multicast library that under the hoods
would appropriately choose ASM, SSM, SSM+hub/spoke distribution,
AMT or worst case unicast replication. Strangely enough nobody from
any university ever stepped forward and said they'd be willing to implement
this and put it into some reference app like vlc or the mbone tools.
Shouldn't be a problem to get such a work funded.

Cheers
    Toerless

_______________________________________________
MBONED mailing list
MBONED@ietf.org
https://www1.ietf.org/mailman/listinfo/mboned