[OLSR-users] Re: routing if hna (0.0.0.0) is < 3 hops and host is > 5 hops
Wed Jul 13 10:27:04 CEST 2005
each of our gateways uses openvpn to a central (cable and wlan connected)
border gateway. IMHO this is how it should be done.
on the gateway:
route delete default gw
route add -host bordergwIP nexthopIP
on the border router:
start openvpn for the gateway
using tha cable connection while still maintaining a coherent IP space.
On Wed, Jul 13, 2005 at 09:03:18AM +0200, Andreas Tnnesen wrote:
> Hi Andrew,
> While your idea is interesting, there are multiple problems:
> - We have no way of knowing if a gateway is implementing NAT. That
> means that we have no way of knowing if it supports "real" routing
> into the manet even if the addresses used are global. Therefore host3s
> gateway might not support "reverse routing".
> - We have no way of knowing what router incoming traffic will be routed
> by into the manet. The traffic to host3 might very well be routed back
> via the GW host1 sent it, or even worse, by a gateway further away.
> - To make sure we can control routing back into the manet we must do
> something along the lines of subnetting within the manet. This means
> nodes must configure/reconfigure IPs based on the nearest GW. Not very
> easy - it would cause much more damage then good IMO.
> However, in IPv6 nodes might configure themselves with one IPv6 address
> pr. gateway. This way a sender can in a way control by which GW packets
> are supposed to be routed back into the manet. Still this assumes that
> a connections dest address is dynamically changed when the topology
> changes which would probably cause TCP issues AFAIK. Also this change
> would have to be transparent to applications...
> Anyways, my reaction is that such a solution is very difficult to
> design if you would want a generic solution and not a very
> special-purpose solution for a spesific scenario/network.
> At least that's my 0.02$ :)
> - Andreas
> Andrew Hodel wrote:
> >Also should note that the implementation should check for a routable
> >IP, i.e. not RFC1918.
> >On 7/12/05, Andrew Hodel <(spam-protected)> wrote:
> >>I have a suggestion that seems would be best in my scenario.
> >>Consider a network with multiple hna nodes to the internet (0.0.0.0).
> >>I want to ping host3 which 6 hops away from me. I am 2 hops from an
> >>HNA node and host3 is 1 hop away from another HNA node. Perhaps we
> >>could configure nodes to route these packets over the internet (or
> >>whatever hna to hna is) instead of taking a 6 hop route directly over
> >>the mesh.
> >>This would of course be a parameter you configure on the hna nodes.
> >>It seems this would effectively erase all routes greater then X number
> >>of hops if other nodes were closer to another HNA node?
> >>On each HNA node you could have a link cost parameter which is
> >>relative to the cost of using a 6+ hop path on the mesh, or going
> >>directly over the internet...
> >>This would sure help routing table sizes in mesh networks...
> >>Please comment, improvise :)
> >olsr-users mailing list
> Andreas Tønnesen
> olsr-users mailing list
More information about the Olsr-users