Re: Resolving geo discussions

[Sean Doran <smd@ab.use.net>]

WRT geographical address assignments: mutually-hostile
sites overlap physically already, and the incidence of this
is increasing.  Site multihoming approaches which do not
take this into account should already be considered unworkable.

Even perfect GPS knowledge in every device does not
eliminate the need to have a DHCP dialogue or some other
configuration activity in the host or in the network (or both),
in the presence of multiple overlapping administrative domains.

Geographical addressing strategies of any variety must
be able to cope with this sort of thing (think of a set of overlapping
802.1 networks at an IETF venue: one for normal hotel
guests, one for hotel operating activities, and one for
IETF attendees, each with different access and usage policies,
wireless network names, passwords, and inter-network
connectivity), or they are simply uninteresting.

On Tuesday, Apr 15, 2003, at 22:14 Europe/London, Iljitsch van Beijnum 
wrote:

> The way I see it, if we start giving out GAPI addresses now, we have a 
> few years to come up with good aggregation before we really need it. 
> If it turns out that geo aggregation is unworkable a solution like 
> MHAP that can use PI addresses without the need to have them in the 
> routing tables can still save us without renumbering. Only if that 
> doesn't work either the GAPI users need to renumber.
Single IPv6 numbers are not obviously rich enough
to convey information about more than one network attachment
point.  Consequently, things with multiple attachment points
need more than one number (leading to a possible explosion
of addresses in the presence of very rich connectivity),
or distant parties need to be able to map single numbers
to multiple attachment points (leading to a possible explosion
of information in the mapping system), or distant parties
must remain ignorant of the existence of multiple attachment
points (leading to non-optimal routing).

Where there is a mechanism whereby numbered
things acquire their numbers in a more timely and
shorter-term fashion than today's long-term static
assignment to DHCP servers and to some hosts,
then the question of what the numbers are
becomes relevant in analysing the risk of these
explosive increases of connectivity information.

Imagine that we could build a GPS receiver into every single device
that is IPv6 capable.   This does not provide these devices insight
into their network locations, because in a single cubic metre
in a telehousing facility, a device's next-hop might be able to
attach to several providers simultaneously, but it might not
be able to attach to all of the providers at once due to failures,
economics, or policy.    Simply knowing where the device
is physically does not allow a  distant party to make a
good decision about which of several different providers'
first-hop routers is the best next-hop to that physical coordinate.

This is already not a purely theoretical problem.  There are already
many individual cubic metres (and centimetres and even millimetres)
wherein there are growing amounts of numbered things.
Some of these things really do need multiple numbers rather
than clever ways of deriving physical location, particularly 
considering,
for example, virtual routers (e.g. cosinecom.com) participating in
multiple discrete virtual and actual network topologies within
different administrative domains.   This is one way some providers
support VPNs, and there are existing real-world cases of deploying
a virtual router platform into a telco central office.   Other 
providers are
offered a virtual access router which in effect behaves and
is managed as if it were a real physical access router in that facility,
as opposed to being offered ATM or ethernet based aggregation
and backhaul.

If you are doing DHCP or configuration activity in the host,
then you might as well use standard topologically assigned ("PA")
addressing and take advantage of the fact that this just works
for the singly-numbered case, and it is easy to acquire multiple
numbers this way, should the host and the site both support that
and find it useful.

There is so little to be gained for the cost of maintaining
a fine-grained geographical address distribution strategy globally
that, to put it bluntly, nobody who is actually in a position to
support such a thing is likely to be remotely interested in doing so.

> On the other hand, if the IRTF comes up with something good
"The IRTF" is us.   The people overlap is substantial.
There is no magic resource hiding in the IRTF that is not here as well.

>  renumbering is a huge pain.
This is the root of the problem.   "Never renumber" is the wrong 
solution.

	Sean.