[OLPC Networking] Re: Ability to set the 802.11s TTL for outgoing packets

[Michail Bletsas]

> 
> > > in the range of 4-5. Is this true? What about the degree of the the
> > > vertices (in the sense of graph theory, i.e. the number of connected
> > > other vertices)? What is the maximum degree of vertices that is
> > > expected in real-world networks? What are the parameters the Marvell
> > > hardware has been designed for?
> > 
> > The number's I've heard are around 2000 nodes, and I have no info
> > about the degree of the vertices.
> 
> Could you point me to a guy who might know this? 
> 
> Or does anyone else on this ML have any idea?
> 
Given than 802.11s uses a DV protocol, the number of nodes in any mesh 
cloud will be limited by the RAM on the 8388.
Besides the routing tables, mesh nodes will have to maintain bridge tables 
when acting as mesh portal nodes.
On top of that one has to factor in the routing control traffic  and the 
application requirements.
So if all you want to run on the mesh is simple instant messaging, you can 
have a few thousand nodes in the same mesh.
If you need to do web surfing, numbers <1000 are doable and numbers <100 
are more realistic for good response times.






> There is another variable in my simulation that I don't know much
> about. That is maximum amount of traffic I am allowed generate on each
> link for my service discovery. The Marvell chip is 54Mbit, right? I
> read somewhere that OLPC will use it in 2MBit mode only, to save
> power. Is that correct? Do yo know how much of these 2Mbit will
> practically be left for application use? I am assuming that I may use
> 10% of that at max. Yes, I know that this value is probably not useful
> without knowing how many machines are part of a the cell, but if i
> know the maximum degree of the vertices I'll know that too.

The usable bandwidth of any specific link depends on the physical 
characteristics of the link itself (distance and noise).
A WiFi link at 54Mbps will allow for about 22Mbps of TCP/IP throughput and 
is guaranteed to work only when the machines are a few (<10) meters apart 
and in clear line of sight. In most cases we expect the links to operate 
at much lower signaling speeds (please notice that in WiFi the maximum 
application throughput is only slightly above 50% of the signaling speed). 
So in your simulations you should factor in a distribution of speeds, from 
the classroom scenario of 22Mbps to the village scenario of 1Mbps (I wish 
I could take the signaling speed even lower for the sake of robustness) 
and you should tilt the distribution heavily towards the lower speeds.


Michail
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.laptop.org/pipermail/networking/attachments/20061010/3e6ab28f/attachment.html