[olpc-software] "Mesh" Software

[Fabian Deutsch]

Hey,

Am Dienstag, den 18.04.2006, 11:02 -0300 schrieb Jim Gettys:
> On Mon, 2006-04-17 at 22:00 +0200, Fabian Deutsch wrote:
> > Hey,
> > 
> > I just looked through the archives but didn't find anything about the
> > software, which is going to be used to create the wireless-mesh.
> > 
> > Maybe 802.11s would have been a good start, but doesn't seem to be ready yet
> 
> I believe Michail Bletsas is planning to use the drafts as a starting
> point for what he's doing in that area.  But as you say, it isn't a
> standard yet, so the best we can do is an approximation, and get to
> discover problems with the proposals in advance.

Yep. We should keep an eye on it ..

> > ...
> > 
> > There are protocols like AODV and OLSR, but I'd personally prefer something
> > more standard. Something like OSPF.
> > 
> > OSPF isn't suitable for wireless meshes currently, but there are some
> > efforts to extend OSPF v3 [1] in a way, so it can be used in MANETs.
> 
> IIRC, Michail says there are 3 routing protocols in existence used in
> meshes; we do not know yet which will be most interesting in which
> circumstances.  I don't know if any of them resemble OSPF.  Which is
> most useful in what circumstances will require more serious
> experimentation.  

Yep. There are some 3 (wikipedia says 70+) protocols, but non of them
seems to be extremly better than others. And some protocols, like TORA
and HSLS are specified, but not found in the wild (I couldn't find any
[community driven] meshes ..).
There will have to be some case studies ...

Because experimantation can get difficult, if you don't have 100+
laptops, you might use some simulator to see some protocols in
action ...
There is ns-2 (http://www.isi.edu/nsnam/ns/), which supports simulation
of olsr, aodv and .. isis (similar to ospf) ... and gtnets
(http://www.ece.gatech.edu/research/labs/MANIACS/GTNetS/) for (ospf,
ospf-mdr)

> 
> The fundamental difference between what we are doing and, say, MIT
> roofnet, is that the nodes are much more mobile than in a fixed system.

True.
OSPF doesn't behave very smart in such environments - yet. But it seems
as if AODV and OLSR are quite optimized for such (quickly changing)
scenarios ... OSPF-MPR/MDR will be.
OLSR seems to be quite popular anyway ...

> 
> A second topic, which is probably worth one or more PHD thesis, is for
> the mesh to be power aware.
> 
> For example, you would probably not want all nodes forwarding all the
> time, if you don't need to, and would want to elect nodes that have
> power preferentially from those that are on battery.  And for those not
> actively forwarding, you'd prefer to keep the radios off for most of the
> time.

I supose that it woul dbe good to use a standard protocoll and tune it's
parameters (path costs, ...) by using plugins, which evaluate system
parameters like "on-battery" and signal strength.

I'd do it this way, 'cause we can rely on some propper protocol
(including algorithms, security ....) and just tune it for wireless use.

(That's in the direction which ospf-mpr/mdr goes - standard based, but
tuned for wireless).

> 
> The Marvell wireless chip selected has enough smarts that such behavior
> could (if one is careful about space use), be done without the main
> CPU's intervention, by having the processor in the Marvell chip turn on
> the radio periodically and see if it should participate.
> 

I'm really looking forward to the first software for this, I can't
really imagine how this is going to work and how this will work with
some kind of routing .. 'cause this is more than just forwarding ...
init?

> > 
> > ANother interesting document can be found here [4], which gives some
> > informations about the behaviour of MANETs with about 100 nodes.
> > 
> > There is already a patch (draft)[2] for quagga[3] and a simulator 2]
> > available.
> > 
> > fabian
> > 
> > --
> > [1] http://www.ietf.org/internet-drafts/draft-ogier-ospf-mdr-position-00.txt
> > [2] http://hipserver.mct.phantomworks.org/ietf/ospf/
> > [3] http://www.quagga.net/
> > [4]
> > http://hipserver.mct.phantomworks.org/ietf/ospf/reports/Boeing-D950-10897-1.pdf
> > 
> 
> Thanks for the pointers.
> 
>                                 Regards,
>                                      - Jim
> 

fabian