Mesh Networking

The mesh networking component in CeroWrt has two components - AHCP, which hands out IP addresses, and babel - which manages determining the best routes available. A full blown mesh node can easily switch between multiple wired and wireless interfaces without losing long term tcp connections (notably ssh), and remain accessible via all interfaces via basically a fixed IP adddress.

By default, in order to co-exist with existing IP address distribution schemes, full mesh capability is not available, however, even partially implemented, it’s useful.

The examples below assume that you are on the default IP addresses supplied by the router. Secondary nodes should have their entire subnets be different than the other nodes.

Mesh networking in cerowrt is not limited to babel and ahcp. Other mesh networking protocols are available as options, notably batman and olsr. In fact many other routing protocols are supported too, all the way up to and including BGP. The main rationale for babel and ahcp is that they are (unlike batman) layer 3 routing protocols and it is much easier to understand what is going on with those than layer 2 protocols. Some things are easier with layer 2 protocols, some are harder, all are interesting ….

Configuring mesh routing on the router(s)

The CeroWrt routers are configured as server mesh nodes by default. This may change in a future release.

You will need to set at least one router up as an AHCP server in order for clients to obtain an ip address on the babel SSID. If they are connected via other means (wired, or wireless STA mode) the routers should have unique IP address ranges, and NAT disabled.

On the server machine:

Change /etc/config/ahcpd to have:

        option 'mode' 'server'
        list 'prefix' '
        # list 'prefix' 'fde6:20f5:c9ac:358::1/64'
        # Usually when I have a /48 to deal with, I set aside something like
        # the:fourtyeight:prefix:bab5::1/64 for ahcp 
        ## You can also setup a real or 6to4 ipv6 address prefix
        ## and/or use site-local ipv6
        ## list 'name_server' 'fde6:20f5:c9ac:358::1'
        list 'name_server' ''
        ## list 'ntp_server' ''

You can have ahcp just distribute ipv6 addresses if you so desire. (this is useful when you only have a /64 to deal with, and in that case you have to disable radvd entirely and all ipv6 nodes have to run ahcp and babel to see each other)

Change /etc/config/network to look like this (or just let ahcp assing these with proto “ahcp”, which is the default.

config 'interface' 'gw11'   
        option 'proto' 'static'     
        option 'ipaddr' '' 
        option 'netmask' ''

config 'interface' 'gw01'   
        option 'proto' 'static'
        option 'ipaddr' ''
        option 'netmask' ''

If you are using babeld: If you are not using the default IP address ranges, you will need to modify /etc/babeld.conf to propagate the routes to those IP address ranges too.

If you are using quagga: the default gw to the internet needs “redistribute kernel” and “redistribute static” lines added to /etc/quagga/babeld.conf. In the case of 6to4, you may need to add a ipv6 default route to zebra.

Restart networking and ahcpd. In the ahcp configuration file, you created a lease address range, starting one above the base number, In /etc/config/network, you will note that the IP address masks in are /32 masks, and the ip addresses assigned are that same base number. The babel protocol figures out where to send data based on ip/interface tuples in this case.

On the client routers (or a server router, actually)

  • YOU WILL HAVE TO RENUMBER. Choose a unique subnet number per router.
cd /etc/config
sed -i s#172.30.42#172.30.SOME_UNIQUE_NUMBER#g * ../babeld.conf /etc/chroot/*/etc/named/*/*
sed -i s#42.30.172#SOME_UNIQUE_NUMBER.30.172#g /etc/chroot/*/etc/named/*/*

Note: the above script is also useful for renumbering a server router.

You will definately need to reboot after running that script.

Change the configuration to be client on /etc/config/ahcp (comment out everything else)

The client routers will automatically pick up an address from AHCP assuming at least one interface is on the same wireless channel as a server,
and then route via the best route to the other routers. Also, get ntp and name service from the right place.

Note that you can just configure all your routers as servers on unique ranges, serve up ahcp from all of them, and still have routes distributed via babel.

Client machines can then just wander from node to node.

Using babel routing without ahcp

There is no strict need to use AHCP to distribute IPs so long as your address ranges are disjoint and babel is told what of the routers addresses to distribute. You can, in that case, dynamically assign addresses via dhcp - but make sure client routers have disabled NAT.

Distributing an ipv6 default route via the main server via quagga

In quagga, there is an ongoing bug (#406) a workaround is to add lines like this to /etc/quagga/zebra.conf

interface 6to4-ge01  ! the name of the 6to4 interface in this case
ipv6 route ::/0 2002:c058:6301:: ! the 6to4 default gw

Other ipv6 mechanisms are similar, but you have to set the default gw and device correctly!

Using ad-hoc mode under wireless on a linux laptop (simplest possible config)

iwconfig the_interface channel XX mode ad-hoc essid babel

I note that many older 5ghz machines do not do ad-hoc properly, and 2.4ghz needs to be used instead. Using ad-hoc mode, rather than managed, gives the ability to move between the highest performing local access points.

So as an example for 2.4 ghz, assuming you are on channel 11 on the mesh router too, and you’re not running Network Manager on this interface

ifconfig wlan0 down
iwconfig wlan0 mode ad-hoc channel 11 essid babel # some devices require this in another order
ifconfig wlan0 up
iwconfig wlan0 mode ad-hoc channel 11 essid babel
# you can also toss these in the background using the -D option
ahcpd -L some_log_file wlan0 &
babeld -z3 -l wlan0 &

so if you get an address, you should then be able to see tons of routes via:

ip route
ip -6 route

Distributing your own default route

Routers that are connected to the internet can and should distribute their default route via uncommenting the default route in the /etc/babeld.conf file.

Mildly different procedure (as noted above) for quagga-babeld

Co-existing with DHCP

Most network managers assume you can have the wired or wireless network connected. AHCP + babel assume you can have both connected. AHCP and DHCP can co-exist to some extent.

Furthermore most dhcp servers distribute a default route, which needs to be disabled in order to failover properly. With ISC dhclient, that involves changing the /etc/dhcp/dhclient.conf file
to not request the ‘route’.

I turn off network manager in Linux entirely, because of these assumptions. You can leave it on if you want to co-exist with dhcp, but it helps to be running babeld and disable getting the default route, thusly:

Change from:

request subnet-mask, broadcast-address, time-offset, *routers*,
        domain-name, domain-name-servers, domain-search, host-name,
        netbios-name-servers, netbios-scope, interface-mtu,
        rfc3442-classless-static-routes, ntp-servers;


request subnet-mask, broadcast-address, time-offset, 
        domain-name, domain-name-servers, domain-search, host-name,
        netbios-name-servers, netbios-scope, interface-mtu,
        rfc3442-classless-static-routes, ntp-servers;

You can also remove domain-name-servers and ntp-servers if you are providing that information via ahcp.

Firewalls, NAT and babel

NAT and Firewalling is a pain in the arse as it breaks the end to end principle. Any place where you don’t need a firewall or NAT (like on an internal router)
TURN IT OFF. the easiest way to disable firewalling on an interior router is to change all your policies in /etc/config/firewall to ‘ACCEPT’.

config defaults
        option input 'ACCEPT'
        option output 'ACCEPT'
        option forward 'ACCEPT'
        option tcp_ecn '1'

To disable NAT change the masq and clamp mss settings in that file to 0.

If (and only if) you disable nat and firewalling and want the external wan interface to be part of the mesh network, you will need to add the ge00 interface to /etc/config/babel (or for quagga, uncomment the relevant line in /etc/quagga/babeld.conf)

I have run into trouble with the default firewall even as it is. It really, really, really wants to firewall stuff.

I have ended up slamming this into /etc/firewall.user on some versions of cerowrt

for i in gw11 gw01
for iptables in iptables ip6tables
$iptables -I FORWARD -o $i -j ACCEPT
$iptables -I INPUT -i $i -j ACCEPT
$iptables -I OUTPUT -o $i -j ACCEPT

A laptop configured via dhcp

The complexities of the babel protocol are hidden in the router, there is no need to run either ahcp or babel on any client machines. I personally like to run babel so I can see the health of my network on the laptop with a tool like babelweb - here’s the mesh network at bloatlab 1 for example.

A laptop configured to be a partial mesh node

You run babel across all the interfaces you have, and ahcp across only the ones you want via ahcp. For example, in my case I have two radios in my laptop, the radios use different channels and get their addresses via ahcp, so as I move around, the best is chosen for routing…

A laptop configured as a full mesh node would run

(Both of these can be configured in a conf file)

ahcpd -d interface interface interface # to get ip addresses on each 'interface'
babeld -z3 -l interface interface interface # to get routes across those interfaces

In this case the laptop also can and will act as a router itself.

Using mesh networking fully

You can serve AHCP across wired interfaces too. This makes moving between wired and wireless modes entirely transparent. However, client nodes also then need to run babel to get routes, as no default route is distributed (default routes are evil).

Pure mesh networking


Ya know, this used to be so much easier before I put in the backward compatibility stuff and just set these boxes up as a pure mesh backbone (as were my old nanostation M5s). You plugged it, you waved it around, and you were online.

Retrofitting it into cerowrt, will require a uci script to convert all the interfaces to AHCP only, among other things.

I want one button on the gui - make me a pure mesh node.

To edit this page, submit a pull request to the Github repository.
RSS feed

Recent News & Articles

Dec 3, 2022 Wiki page
Codel Wiki
Jun 11, 2022 Wiki page
More about Bufferbloat
Jun 11, 2022 Wiki page
What Can I Do About Bufferbloat?
Jun 11, 2022 Wiki page
Tests for Bufferbloat
Dec 7, 2021 Wiki page
Getting SQM Running Right

Find us elsewhere

Bufferbloat Mailing Lists
#bufferbloat on Twitter
Google+ group
Archived Bufferbloat pages from the Wayback Machine


Comcast Research Innovation Fund
Nlnet Foundation
Shuttleworth Foundation

Bufferbloat Related Projects

OpenWrt Project
Congestion Control Blog
Flent Network Test Suite
The Cake shaper
CeroWrt (where it all started)

Network Performance Related Resources

Jim Gettys' Blog - The chairman of the Fjord
Toke's Blog - Karlstad University's work on bloat
Voip Users Conference - Weekly Videoconference mostly about voip
Candelatech - A wifi testing company that "gets it".