[olsr-dev] Combating hidden nodes using token-passing

[Dan Flett]

Hi All,

I've recently taken an interest in OLSR networks after a long time of being 
a staunch advocate of infrastructure-style wireless networking.

Some of us in the Melbourne Wireless group have been tinkering with the 
Frottle package
http://frottle.sourceforge.net/
developed by WAFreeNet.

It uses a master/client polling method to overcome the hidden-node problems 
that occur with the 802.11 CSMA/CD media access method.  It uses the 
Iptables userspace QUEUE module to hold back traffic and release it only 
when polled by the master.  It's a Layer 3 solution to a Layer 2 problem and 
it works well - and being layer 3 it is chipset agnostic - it works with any 
wireless card.  It even works with every stand-alone Access Point in 
existence if you connect it to a Linux box.

Recently we were contacted by a wireless group who operate in the Himalayas 
- they are using OLSR and they are having hidden-node issues with their mesh 
network.  As you can imagine - with some nodes on top of huge mountains and 
some in valleys, a good portion of the nodes in the network are hidden from 
each other.  They are using Frottle, but it's current architecture limits it 
to one master node per subnet - nodes that are more than one hop from the 
master can't effectively participate in the network.

We are beginning to think about a way of using Frottle on a mesh network.  
It currently is only really suitable for an infrastructure-style network 
with an AP that polls it's clients to transmit in turn.  We've decided that 
a token-passing media-access method is the way to go - and it turns out this 
method has been the subject of a lot of academic research for some years 
now.

Some Googling turned up these papers:
http://www.eecs.berkeley.edu/~ergen/docs/wtrpiscc.pdf
An overview of a Wireless Token Ring Protocol - a good intro, but lacking in 
detail

http://faculty.cs.tamu.edu/welch/papers/icnp01.pdf
Distributed Token Circulation on Mobile Ad Hoc Networks
This paper reviews the different algorithms used to pass a token around all 
the nodes on a mesh network.

Some of the more efficient methods for passing tokens around require that 
each node have a good idea of the overall topography of the mesh network - 
and this, of course, is information that is gathered by OLSR.  However, a 
program like Frottle will only work properly if it has absolute control over 
when data is transmitted - OLSR transmissions would have to wait until 
Frottle is polled (or receives the token).  So a mesh version of Frottle 
will have to send out it's own Hello packets to discover other nodes - at 
least in the inital discovery stages.

At some point we are going to have a go at developing a token-passing 
version of Frottle.  We're still at the thinking stage at the moment.  We 
want to get a basic, stand-alone version working first.
The ultimate goal is to increase the efficiency of the token-passing using 
the data gathered with OLSR.  We are interested in hearing from anyone who 
is interested in, thinking about, or working on a project like this.

Cheers,

Dan Flett