[OLSR-users] More bridge weirdness

[John Gorkos]

OLSR users-
  I'm finding more weirdness with my wireless bridge setup.  I refer you again 
to a current snapshot of my mesh network:
http://clearviewcity.wildcatwireless.net/cgi-bin/post_mesh_once.pl

the 192.168.7.1 node is the only takeout point.  In the olsrd.conf, it has 
this line:
HNA4 0.0.0.0 0.0.0.0
and it is the ONLY router in the network with this line.  However, as I 
mentioned in a previous post, the .7.1 machine is a standard (x86) linux box 
with two ethernet cards.  eth1 is 192.168.7.1 and is connected directly to a 
smartBridges airBridge outdoor wireless bridge , which is in 
point-to-multipoint bridge mode.  It has an ip of 192.168.7.2, but should be 
transparent to most IP traffic.

Currently, all nodes have a VPN tunnel to 192.168.7.1 which they use for their 
default route.  Unfortunately, it looks like openvpn does not really fare 
well with routes that shift as quickly as they seem to in a mesh network, and 
the openvpn daemons tend to lock up.  This makes my customers call me with 
the common complaint, "The internet is down."  After trying to explain to the 
first few that the internet was, indeed, still working fine and only their 
CONNECTION to the internet was down, I gave up and just started rebooting 
routers.  Invariably, the simple act of rebooting a WRT54G would make the 
global internet spring back to life, and my customers could get on with their 
cyber lives.

Lately, I've been experimenting with NOT using the VPN tunnels and simply 
letting olsr do it's thing and route packets off the network.  After more 
than a week of pounding my head against iptables, manual routes, tcpdump, 
etc, it seems that something is simply preventing packets from getting to the 
gateway (.7.1) machine.  Furthermore, take the following simplified mesh:

   .25                .1                   .20
     x-------------x---------------x
      |                /   \                  |
      |               /      \                |
      |             /          \              |
      |           /               \           |
      |         /                   \         |
      |      /                         \      |
      |   /                                \  |
       x                                   x
     .31                                 .27

doing ping tests, it seems that all nodes can ping .1
.25 can ping .31, and .20 can ping .27
BUT, neither .25 nor .31 can get to .20 or .27, and neither .20 nor .27 can 
get to .25 or .31.  It seems that my gateway simply refuses to play router to 
move packets from one side of the network to the other.  It DOES act as a 
good VPN server, and all packets arriving on tun0 get masqueraded and moved 
off the network appropriately.
The only thing different about .1 is that it is an i86 machine behind a 
bridge, whereas everything else in the network is a WRT54G using the built in 
radio (eth1) for the OLSR interface.  I have completey purged all firewall 
rules off the .1 machine, save a single masquerade POSTROUTING rule so I can 
stay in business.
Questions:
  Does anyone else have experience running olsr through a wireless bridge, and 
if so, is it successful.
  Can anyone think of something I may have missed to explain why .1 refuses to 
route packets in the mesh (yes, /proc/sys/net/ipv4/ip_forward=1)?

Finally, what's the biggest "real-world" mesh in terms of a) number of nodes 
and b) geographic area that anyone is running with UniK-olsrd?  I'm starting 
to wonder if I'm moving into uncharted territory.  My client has installed 
all but 1 of the initial 20 routers, and I'll be ordering the second set of 
20 next week, and interest in the apartment complex is really picking up.  
I'm ready for advice on how to scale this beast.

Thanks,
John Gorkos
Wildcat Wireless