Node Mobility in o11s

[Javier Cardona]

Hi Mike,

On Mon, Oct 13, 2008 at 10:06 PM, Mike Kuredijan
<michael.kuredjian at gmail.com> wrote:
> After reading Keir's message to the list in October archives, I have
> to express relief that there are others out there attempting to do the
> same as I :) I've got a few test machines available, running
> 2.6.27-rc6-wl and two BCM4318-based cards. I haven't installed the
> cards into these test machines just yet, but I plan to do so tomorrow.
> One problem I did come across when running both cards in one machine
> is an inability to create more than one mesh interface. Attempts to
> create a second from the next device results in an error -23.

This is a general mac80211 limitation, not just mesh.  Each mode of
operation (infra, ad-hoc, mesh) configures the physical layer in a
different way.  For instance, infra will turn on BSSID filtering if
available, and mesh requires that filter to be turned off, so both
types of interface cannot coexist.

> My aim to is do more digging into kernel routing, but what kind of node mobility testing were you thinking of?

Designing tests that simulate typical mobility patterns of pedestrians
moving around a mesh network is not easy.  But there are simpler tests
that can be done to measure the healing time of a mesh network.

One such test would be to build a mesh network with redundant nodes
(i.e. so that there are not cut vertices in the resulting graph), and
see how pairwise connectivity and throughput are affected by randomly
disconnecting nodes.

Another interesting test would be to vary the transmission power in
random nodes to simulate movement.  This is not exactly the same as
moving nodes apart (unless the receiver sensitivity can also be
adjusted) but should trigger changes in mesh paths.  This tests should
reveal how well the path selection algorithm deals with same cost
paths.

Cheers,

Javier