[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: I-D ACTION:draft-ietf-v6ops-nap-01.txt
Hi Fred,
On Fri, 1 Jul 2005 08:25:03 -0700
Fred Baker <fred@cisco.com> wrote:
>
> On Jul 1, 2005, at 4:03 AM, Mark Smith wrote:
>
> > (2) Running a fully fledged routing protocol on the end-node is
> > relatively heavy handed for what is being achieved. I'd think a
> > simple, light-weight announcement style availability protocol for the
> > node's address, that is part of IPv6 itself, would be useful.
>
> It was standard practice on DECNET, XNS, and Novell equipment; Novell
> did it twice, once for the routes and once for the services that the
> servers offered. If you have only one interface, as you say it didn't
> accomplish much. But if you have multiple interfaces on a server, it
> provides something akin to what multihoming does for a network. As you
> say, it is expensive in address count to put a /64 on a host, though.
>
It took many years for me to realise why there was an "internal network
number" on Novell servers :-) I think it probably co-incided with me
learning how /32s on loopbacks pushed into a routing cloud could be
useful for maintaining routers.
> In general, if you're interested in routing scaling, using the tools
> that make it scale are a good thing. Host routes are the exception, I
> should think, not the rule.
I generally agree, I personally don't really see all that much security
value in attempting to make the internal subnet topology opaque to
external devices at the cost of scalabilty. However, people seem to
like NAT because that is one if its properties, and I could see them
wanting to preserve it with IPv6. The NAP draft is suggesting
using host routes to do this, and how that will work in IPv6 has
intersected a bit with my somewhat academic interest in how it works on
broadcast interfaces with IPv4.
For those who do place value on topology hiding, I'd imagine it would
most likely to be "all-or-nothing" approach, meaning that all their
hosts within their assigned /48(s) would be host routed, rather than a
minor subset. If they think topology hiding is important, they're
probably also prepared to think it is important for all their nodes,
rather than a minor subset, and be willing to wear the scalability cost.
Regards,
Mark.