batman-adv...

[rebelwireless]

Ok, I know this sounds crazy but hear me out.

I have a test lab where I'm using APU 2C4 units running debian and batman-adv.

a couple nodes have the following configuration
br0 contains bat0 and eth0
bat0 contains eth1 and eth2
effectively, eth0 is access and eth1,2 are mesh.

other nodes just have bat containing eth0,1,2.

I'm able to seamlessly connect across 10 APU 2c4 boards via any access port and convergence for failed 'paths' is very quick.

Further, I've attached ubiquiti UAP-AC-Lites to 3 'access' nodes and have a 4th hooked up to the 'wan' network. UAPs think they are just connected to a switch. I can seamlessly roam between APs even though these are connected across the mesh network.


This configuration is truly seamless. To complicate it, I've spun up a xenserver box and have 20 VM's running batman-adv connected to the physical mesh and it's all completely seamless.

What's the point? I want to see batman-adv on the netonix switches. batman-adv can live on top of a VLAN, which means a simple VLAN access port can be a member of the mesh.


This effectively replaces the need for SPB, it will run on anything Linux will run on, and performance can be very good.


thoughts?

[mike99]

Batman-adv is great and I once wanted to use it too but it's only on custom router and the Netonix CPU probably couldn't handle much traffic.

[sirhc]

The Netonix CPU is not very powerful, it is only a MIPS 24K.

Our switch core handles all the packet forwarding at line speed so a 26 port is capable of 26G simultaneous switching. The switch core is independent of the MIPS 24K which is used only to run the UI / CLI / STATS / and some misc daemons.

The switch core has its own specialized processors (many) to handle packet switching/forwarding and other Layer 2 functions. The processors in the core are not accessible by the user.

Reference this post and others:
viewtopic.php?f=6&t=2817&p=19426&hilit=+UI+CLI+MIPS#p19426

[rebelwireless]

batman-adv is more of a ram hog than a CPU hog by virtue of keeping a state table for all known mesh nodes. Of course, it is encapsulating packets on ingress and egress (just forwarding on mesh nodes) so there is some CPU hit and that would limit the switch's capabilities, but very light CPU's can handle quite a bit of traffic with batman-adv.

I'm experimenting with router-on-a-stick with an APU2C4 and VLANs which is working pretty well. Biggest issue is I need to burn 3 ports on the switch to get the maximum out of the 3 gigabit ports on the APU. Not a big deal at a small site where the backhaul is <1Gbps, but a little less elegant at sites with 2+ backhauls.

[Blaha]

Why's it called batman-adv? Someone clue me in please.

[rebelwireless]

Better Approach To Mobile AdHoc Networking. Advanced... There was a 'b.a.t.m.a.n.' protocol that meshed over layer7, the -adv meshes over layer2 creating a seamless switch.

[mike99]

Batman was a protocol that follow up OLSR to fix what was learn from it. Batman-adv was a new approch to batman that use encapsulation to hide the network from user. From whay I understand, Batman-adv is similar to STB, or TRILL, but created for ad-hoc network where everything is encapsulated over a "layer 3" network. You need batman-adv tool to be able to "ping" and "trace" the "route" of data.

It's a really cool project that would probably be great for WISP. Batman-adv is probably packet loss aware like OLSR, but , just like STB, no switch chip availaible to small manifacturer like Qualcom or Microchip (Vitesse) support it, at less, from what I know. It's really a linux based project. I'm not even sure a switch chip would be a good idea for batman since batman normally bridge interfaces via software and would see all switch port as a single device.

[ginovilla]

how its going with the batman on a stick?