[OLSR-users] [WLANware] Selling the idea of OLSR

[Eric Huiban]

Dan Flett a écrit :

>Hi guys,
>
>I am giving a presentation to my wireless group, Melbourne Wireless, this
>Friday on OLSR.  We currently use OSPF and a traditional routed network
>where each node has multiple radios, each with a separate /28 subnet.  Most
>of our links are point-to-point or point-to-multipoint (client-to-AP).  A
>lot of people are very sceptical of the idea of using a /16 netmask on an
>interface and of using Omni-to-Omni links.
>
>Here are some of the criticisms I'll need to address:
>
>* Mesh networks are not scalable - they cause too much CPU load on small
>routers.
>  
>
This is the major remark i've heard from people who sticks to classic 
networking "mind" where they
tend to  have a global image of the network at once. OLSR induce not too 
much CPU load on small
device due to the distributed  calculation of routing. None of the 
routers has to compute the whole routing
and nobody cares about it : only one and two hop routing is computed on 
each node and this is not
too much complicated even for a very light device.

It only becames complicated for people (big-brothers, bofh, etc) who 
want to master every second
of the network life... but for cpus doing their olsr duty, it always 
remains quite simple due to their
limited aerial field of vue. :-)

I noticed one case where the nodes will becomes very busy : the mobile 
network with some OLSR
routers bolted on cars, scooters, etc. Pedestrian routers are less prone 
to induce routes computing overload.
But tuning the beaconing parameters can limit the intrusion of 
"high-speed routers" and can give
more stability to the whole static and "low-speed" network part.

I'm currently woring on a small network in south of France which is at 
its very beginning with OLSR.
But what i saw in town in Germany and Netherland is very promising when 
dealing about scalability
in term of number of nodes.

>* Omni antennas pick up too much RF noise for usable links
>  
>
This only is the sympthom of very badly choosen antenna !

In low density aera you may need two antennas, one dedicated to 
long-jump, often a directionnal one,
and one dedicated to your own local area coverage, often an omni.

About antennas, there is only one truth : feel free to express your 
adaptability and your creativity !

For multiple antennas installation, (even if everything is currently 
only on my bench) i tend to promote
the use of active antennas with ethernet connection directly to the 
router. It's not exactly cheap but it 
don't need any software modification to ethernet/wifi bridges and 
efforts have a best employment when
applied to the router part.

And when node density increases (but there is no none case today), in a 
very crowded aera, you
may have to forget high gain antenna and switch back to the router 
original antenna in order to
reduce your filed of view and ensure a better stability. In overcrowded 
place you may also have
to reduce the TX power...

>* Omni-to-Omni links have limited range and the population density in
>Melbourne is not sufficient for a critical mass to form a useful mesh
>
>  
>
You're not obliged to use omnidirectionnal lins every where. 
Directionnal antennas and
"sectorial" antennas (directionnal but with a greated apareture) give a 
great help when trying
to perform long-jumps with one or several colocated targets. This kind 
of antenna also allows
you "tune" your OLSR behavior without inducing any additionnal filtering 
calculation, simply
by fiddling with physical world. In a few words : just use the antenna 
that you need on your
very own location, mesh is not tied to omni use !

>* A network where everyone is using a /16 netmask will be like a giant LAN
>and everyone will be swamped with everyone else's broadcast traffic
>
>  
>
Like previously expressed people with this remark in mind forget that 
they are not using a wire !
the whole network will not be seen by all nodes. Only one and two hop 
nodes will be seen by an
other node. And only one hop broadcast will be percieved. A whole OLSR 
network can be seen
as  an "cluster" (?) of small OLSR network each one based around each 
participating node.

People learning to be "master of world" are here phased out... OLSR has 
a kind of equity when
dealing with network management. The simplicity and the "no needed 
configuration" allows
people without technical background to participate to the network and to 
drop node everywhere
they want.

There is an other forgoten aspect when dealing with ad-hoc network. The 
transceiver is physicaly
limited to its field of view and cannot go behind it. On a wired network 
you can flood the entire
network but not with hertzian network. your broadcast is limited to your 
neighbour and you will
became a real beloved guy by your neighbourhood  (and only by them) if 
you use broadcasting
application on an hertzian network...

>* Using OLSR on every node is like trying to create a city-wide network with
>WDS
>
>  
>
The main difference is that networt structure will need no tuning of 
OLSR nodes once configured for
one network. It can be dropped everywhere in town by anybody without 
further tuning.

WDS is only an emulation of a wire and has the same installation 
constraints on each end. This is again
a feature for infrastructure minded guys who can't get rid (?) of it by 
obligation while working for acces
provider having a classical infrastructure, or by inadaptation problem 
when speaking about mesh networks.

>I think I have answers for most of these questions, but I'd like to get some
>ideas from the wider OLSR community.  I'm sure many of you have faced
>similar criticisms, and may have some experience behind you to refute these
>claims.  If so, I'd love to hear from you!
> 
>
Good luck!

Regards,
Eric Huiban.