Re: ISP scenarios comments (multicast)

[Brian Haberman <bkhabs@nc.rr.com>]

Pekka,

Pekka Savola wrote:
> 
> On Thu, 19 Sep 2002, bkhabs wrote:
> > >Ok; this is still a bit problematic with the one PIM domain requirement..
> > >:-/
> >
> > I don't think it is problematic when you consider the overhead that
> > would have to occur in order to support PIM domain != site-local domain.
> >
> > The major problem is that PIM messages are mostly sent hop-by-hop with
> > IP addresses in the payload.  So, the Bootstrap message is transmitted
> > hop-by-hop using the All-PIM-Routers link-local multicast address.  But
> > the BSR address is in the payload.  That means that a site-local border
> > router would have to recognize the Bootstrap message, parse into the
> > PIM payload and check to see if the BSR-Address was a site-local.
> >
> > In this case, I prefer simplicity.  Let pim domain == site-domain.
> 
> I agree with the simplicity -- but the problem I was referring to was due
> to the fact that for all practical purposes, currently there can only be
> one PIM domain.  That limits the applicability of multicast quite a bit.
> 

Well, actually you can have more if the PIM domains are completely
contained within a site-local domain.  Using site-locals for multicast
is limiting the applicability anyway.

> For one perspective, look at one PIM-domain at:
> 
> http://sem2.renater.fr/m6bone/
> 
> Multicast works -- kinda.

Yes, I have looked at their network before.  Interesting.  I am
assuming that they are not using site-locals.

> 
> > >This might be an interesting point to develop more.  It's quite possible
> > >if we can hard-code assumptions about address assignment for the RP there.
> >
> > >Possibly through an additional bit flag in the address. This would have to
> > >go to the PIM implementation then, though.
> >
> > You can't encode the RP in the field.
> 
> Which field?  Destination address could include the address of the RP,
> with some rather strict assumptions.

I was talking about the prefix field in the multicast address.  There
are not enough bits to represent an RP unless you did some interesting
address assignments for RPs and can encode the entire address in 64
bits.

> 
> > But, a new RP discovery mechanism
> > could utilize the prefix information to know where to look for the RP.
> 
> That too.

Brian