Re: IPv6, interNAT, Wi-Fi (not mobile)
2003-03-27 13:29:25
On Thursday, March 27, 2003, at 11:29 AM, John Stracke wrote:
S Woodside wrote:
On Tuesday, March 25, 2003, at 06:03 PM, John Stracke wrote:
proponents want to be able to do massive multihoming, with all
participants with external links sharing those links, and all the
traffic from the outside finding the shortest way in. I won't say
it's impossible, but last I heard nobody knew how to do it; the
route flap would be horrible.
That said I understand that it's not a novel challenge. However this
is a novel application AFAICT, and it has some characteristics that
make it potentially different from the scenarios I've seen analysed
(mostly mobile, or fixed but wired). The fixed wireless mesh will
present a planar graph topology more-or-less, due to the inherent
distance limitation of microwave signals. Also, the nodes are likely
to be stable over time because they are mounted on houses or other
kinds of buildings for long-term use.
The set of possible links may be stable, but their presence or absence
won't be; you'll have people unplugging their machines, or rebooting
them, or losing connection while someone's running the microwave.
Understood. But the stability will be much greater than in a mobile
situation. I would expect the AP to be most likely in a separate box,
and it to be acceptable for a few minutes to pass before recovering
from downtime.
Suppose I have no direct Internet access, but I have links to three
neighbors that do. One of them has DSL, so I route through him when
he's up. But DSL seems to get flaky when it's hot (I don't know
why--and I'm not positive it's true, but there are other ways to flake
out, so bear with me); so, come August, his line starts going up and
down every few hours. When it does, I broadcast a route update to go
through my other neighbor's modem line. Now *every* *single* *router*
in the default-free zone of the Internet (the backbone routers, the
ones that don't just connect to a single upstream provider) has to
track my updates. This does not scale. It's *such* a Bad Thing that
backbone providers flat-out refuse to accept routes for individual
hosts--I think most of them refuse anything smaller than a /24.
This is routing flap right? So, in this case it seems like damping such
a flap would be totally appropriate ... and indeed would have a
positive effect on the mesh by encouraging nodes on the mesh to produce
more interconnections to oppose the effects of being cut off from
routing. More interconnections, more interconnections, can have the
opposite effect of increasing the health of the network (modulo routing
protocols of course...)
But routing could be implemented such that the DFZ doesn't need to see
all these routes. Again, I think that the planar-graph nature of the
network comes into play here. Geographic routing can break up the
routing into arbitrary levels of precision, so that packets are routed
to continents, then provinces, then localities, then specific nodes (or
whatever). In this scheme changes would only need to be propagated up
to the level where they stop affecting routing decisions. It's the
nature of the mesh that makes this possible whereas the current wired
infrastructure, there really is no way to know "where" the bandwidth is
going since wires don't discriminate distance -- but wireless
(especially microwave) does.
I've seen geographic routing and various geocoding schemes involving
embedding GPS etc. which is really overkill. Any given mesh segment
need only know it's position roughly, and once packets reach the right
general area they can be routed using a different technique that's
non-geographic if that's more appropriate. In addition since they are
not mobile each node need only be configured once as to its location,
after that there is no change.
simon
<Prev in Thread] |
Current Thread |
[Next in Thread> |
- Re: IPv6, interNAT, Wi-Fi (not mobile),
S Woodside <=
|
|
|