Yesterday I demonstrated a project about which I'd asked some questions here (comp.dcom.voice-over-ip). I haven't yet written up the project but I appreciate the help I got so here's a brief summary.
At the site's office, I installed a panel computer connected to a switch that's fed by a T1. The computer provided NAT service to the wireless network and used a magnetic-mount antenna stuck to the side of the building. (It did *not* work well from inside. I finally realized that the glass was not transparent to RF.)
In the main building I installed a similar panel computer but this one fed another SPA-2000 with two telephones on its wired network.
On a catwalk above some nearby grain bins, I placed another WRT. It operated only as a repeater. Power was from 120VAC although I had batteries ready to use.
All of the machines ran OLSR. Except for the T1 gateway, they provided routing and DHCP to their wired networks and advertised these networks through OLSR.
From the main building I could watch the routes as the wagon moved around on the tour. My advisor and my wife arrived in the afternoon and we finally tested using VoIP and it worked surprisingly well. I'd only used it in my house before this and it's hard to tell how good it is when the other person is close enough to hear without the telephone.
Unfortunately the router got flakey as I tried to test it while moving. I reset it once and it seemed to be o.k. but then it went out again. I finally realized that being wedged up against the window in a closed cab might be a problem. It was *hot* to the touch so I pulled the power.
So...I didn't get to test voice continuity across routing changes but I do have a log of pings to the device which should give a decent idea of how it might perform.
I'd wanted to have another vehicle rigged in order to exercise the mesh some more but instead I roamed the grounds with my laptop and took notes.
Things learned...
Sipura devices are nice because they have a lot of built-in intelligence. I wrote a script to query for all of the connected devices and then set the autodialers on each to contact all of the others directly without registration. Being able to avoid having a full computer for each remote is handy.
Consumer electronics shouldn't be expected to work well when roasted.
OpenWRT is wonderful. I hope to exploit it much more. The WRT boxes are incredibly handy. For events, attaching some to deep-cycle 12V batteries would be a quick way to establish a wireless network.
OLSR is usable out of the box. It allows tweaking though and some thought should be given to using it in more taxing situations. It definitely did some things that didn't seem quite right and was not always solid but I expect that it can be configured to behave in a more desirable way.
I deployed this network in an hour or two. Decent planning, multiple people and plenty of zip ties could mean being able to deploy a network like this in a few minutes. I think about building a system that could be checked out from the university for remote events. It could also be useful in emergency situations.
--kyler