After a bit of a hiatus from FabFiing, I went back to the drawing board on the FabFi architecture, with the hope that I could solve the downstream routing issue and add meshing capability to the mix. The way we're trying to use Fabfi in JBad is special because we want both a long-haul backbone and local mesh using one setup routine. Additionally, the concept of AP and STA has proven confusing for the local users, so deleting it would be sweet...
Since we started on FF1.0, OpenWRT came out with a new release (8.09) that incorporated a bunch of new features, including an integrated web-admin GUI, and functioning OLSR packages, so after playing around with freifunk, which is still stuck back in Whiterussian, I flashed a few routers with 8.09 to see what they could do.
...and they did, but only very very slowly (72kbps slowly).
Here's the setup:
- 3 vlans: WAN, LAN, MESH; Wireless bridged to MESH
- olsr-luci runs on the br-mesh bridge (Wireless and wired MESH)
- The olsr-luci-mod-dyn-gw-plain plugin advetises nodes with net uplinks to the mesh
- The olsr-luci-mod-nameservice plugin passes around DNS server info from the nodes with uplinks.
All of the nodes are connected on the MESH addressed to 10.1.0.0/24. Each node is statically given an IP (last octed is FabFi number). The nodes can be connected wirelessly or by wire.
Any node can provide an uplink, but as of this setup the user must specify that it is acting as such by identifying it as a DNS server.
Every node provides a LAN on 10.1.x.0/24 (x is Fabfi number).
Which nodes wirelessly link to each other can be controlled with the SSIDs and/or wireless channel (only matching ones link).
The Results:
good:
- All nodes link nicely, and seem to do the meshing thing.
- One can ping any node in the mesh, and there are routes available to any subnet.
- All the nodes can see the net
not so good:
- The net is slooooow. (72kbps)
- If dnsmasq is started / restarted after olsrd, the DNS server information provided by the nameservice plugin gets blown away (presumably until the server changes? Not sure)
Ideas anyone?
Links
Subscribe via email
Subscribe via feed
Posts
Comments
Labels
- Antenna Tech (10)
- Bugs patches + updates (1)
- Community Building (15)
- FabFi 2.0 (7)
- FabFi 3.0 (1)
- Fabfi 4.0 (1)
- Fabfi Dashboard (1)
- FabFi Intranet (1)
- FabFi Toolbox (7)
- Field Tests (9)
- Introduction (1)
- Kenya (4)
- Network Configuration (2)
- Network Monitoring (2)
- New Links (8)
- Random Frustration (10)
- Router Tech (12)
- System Development (1)
- WiFi Tech (2)
3 comments:
Is it slow because it might still be converging? Or has it fully converged? Also what's the inter-node layer 1 look like? Good (or great SNR), low jitter & few retries?
Remember it's CSMA/CA control protocal maybe hidden node?
Can you test to see if different through mesh routes are better then others?
Result was the same after letting it converge over night. Result also the same when there are only two nodes in the system.
the wireless is definitely the cuprit. When I disable the wireless on one device and connect the nodes with a wired lan, they're fast enough to max out my net connection (10,000kbps). If both wireless and wired are connected, result is the same as having only wireless.
The wireless signal strength is excellen--SNR -9/-95 (the boxes are right next to each other)--but I notice the excessive retry count is huge. Seems like the timing is off somehow. I'm not sure how to test/change...
btw, out of the box the wireless on 8.09 seems to work well. I get the same speed for wifi as I got for wired above (>10,000kbps, presumably my net connection is the bottleneck)
Post a Comment