Re: Death of the Internet - details at 11

Noel,

(1) Sorry to have misconstrued your comments.

(2) Yes, I was trying to refer to situations in which each of 
the hosts on a multihomed LAN has exactly one address, largely 
because of bad experiences with client machines running 
widely-used junk software trying to handle multiple addresses 
(in IPv4) on the same interface.  I know of only two ways to 
accomplish that "exactly one address on local host" criterion: 
(i) use of NATs (even if only on a one-one basis) to make 
different external addresses appear as a single set of addresses 
on the LAN and (ii) use of globally-routed, 
provider-independent, addresses for the LAN.
I am now going to drop out of this discussion since I don't seem 
to be in sufficiently good shape this week to make comments 
sufficiently precise that I don't just create more confusion.
    john


--On Thursday, 29 January, 2004 11:29 -0500 Noel Chiappa 
<jnc(_at_)mercury(_dot_)lcs(_dot_)mit(_dot_)edu> wrote:
>     > From: John C Klensin <john-ietf(_at_)jck(_dot_)com>
>
>     > Of course, multiple A records works, is out there, and
> have worked for     > years. But they worked better before we
> introduced routers (i.e., when     > the hosts with multiple A
> records really had interfaces on different     > networks).
> Today, it effectively implies having multiple addresses on
> > an interface and multiple "local" address prefixes running
> around on     > the same physical LAN segment.
>     > ...
>     > Perhaps more important, as Noel points out, it doesn't
> scale very well,     > at least in terms of the routing fabric.
>
> Sorry? What I said doesn't work for the routing, in terms of
> scaling to many small sites, is for a multi-homed site to have
> a single address prefix, which is then globally advertised.
> (That's the most common tack for multi-homing support in IPv4
> to date, which is what you were talking about.)
>
> Having multiple addresses for a host (which has only a single
> physical interface, but which is in a site which is
> multi-homed) is in fact the only approach whose effects on the
> routing does scale (within anything like the current routing
> architecture, i.e. packets which include only source and
> destination addresses, as opposed to a source route).
>
>
>     > as I tried to point out, address preservation policies
> have had     > trickle-down effects that make it impractical
> for small enterprises.
>
> It is an interesting point (as Daniel Senie also just pointed
> out) that multiple addresses -> faster consumption of the
> address space.
>
> However, since to avoid a size explosion in the routing
> tables, those multiple addresses do have to be
> connectivity-dependent (political translation -
> "provider-dependent"), I don't see how address preservation
> policies have made this approach "impractical for small
> enterprises". My understanding of current allocation policies
> is that ISP's can get enough addresses to cover their
> customers. If company X is a customer of both ISP P and ISP Q,
> I would assume that both P and Q don't have a problem getting
> enough space to cover their customers - including X.
>
> (Or were you speaking of the "one address block, globally
> advertized"?)
>
>         Noel