Re: The state of IPv6 multihoming development

[Christian Kuhtz <ck@arch.bellsouth.net>]

On Sun, Oct 27, 2002 at 06:50:29PM -0800, Tony Hain wrote:
> Michel Py wrote:
> > ...
> > In short: a large organization needs to be able to engineer 
> > traffic, which I don't realistically see happening when 
> > individual hosts can take decisions that can affect traffic 
> > engineering.
> 
> A host decision does not affect traffic shaping for the outbound
> direction. The only choice it is making in that space is the dst addr,
> and it has that responsibility today, even though it has no
> deterministic means of making that choice. The IPv6 address selection
> rules add structure to that process.

maybe i got this all wrong, but...

if this means that a host is expected to make a decision which path to 
chose, i'm sorry, i don't see that happening and consider it an absurd 
notion.

the last thing i want is for an oracle server, or webserver, or whatever
to make a routing decision as to which addr or iface to use.  seems the
trend is exactly the opposite.. to get that decision _away_ from the host
and have the backbone of a given org take care of traffic management.

any type of te will have to aggregated at a higher level than hosts. 
 
some of this is a technology and architectural decision, some of this is
also real world organizational skills.  i do not want to create a rqmt 
that says that systems operations folks have to be omnipotent systems &
network wizards to configuration, manage and troubleshoot their hosts.

and i think that's exactly what you do when you push te capabilities down
to the host level.

> For the inbound direction, the remote and intermediate network managers
> have more impact on the actual path the bits will take than the origin
> choice of source addr. All we have today is the illusion that a network
> manager can direct traffic toward his network. Having 1 or N addresses
> doesn't change the granularity of traffic engineering, so the argument
> is bogus. What it really boils down to is that nodes outside the network
> manager's direct control are making a decision (any decision). This is
> about trust, not traffic engineering.

nevermind that there are various bgp knobs today which allow you to do
exactly that (regulate inbound traffic) by changing what prefixes are
advertised where and how.  there is a gating factor here in today's ipv4
world in terms of the smallest chunk of addresses which can be pulled out
of a given aggregate.

thanks,
christian