Scalability of manets (was: Re: [OLSR-users] Question about OLSR

[aaron]

On Wed, 12 Jan 2005, Jens Nachtigall wrote:

another point on scalability.
The discussion so far failed to mention the distribution of load which I 
do hope occurs when you employ LQ under the following assumption

If you assume that one node has a few (dedicated, i.e. directional) links 
to other nodes, then the whole graph could represent a max flow 
problem. This is well researched in literature.

Correct me if I am wrong...

Also:
There is a company in the states (i believe meshnetworks.com or 
so?) who claims to have solved the scalability issue. Anybody know any 
details?

ciao,
aaron.


> Dear all,
>
>>> your
>>> bandwidth will be saturated by control messages alone.
>>
>> Not only that, but you must remember that standard WLAN today only
>> have one antenna, so it cannot send and receive at the same time.
>
> Actually it is even worse :-(. Afaik the biggest scalability problem is
> not the one of increased control messages, or interference as outlined
> above. But the biggest problem is, that if the net increases my node
> will have to forward more and more packets on behalf of other nodes.
> Therefore, my node can hardly issue packets on his own.
>
> If the traffic does not remain primarily local when the manet grows,
> then the throughput available for each node will grow towards 0.
> I don't want to go into detail here, but there is a very nice paper on
> http://www.pdos.lcs.mit.edu/papers/grid:mobicom01/paper.ps.gz
> (8 pages, not too much brainfuck). The paper sums scalability issues up
> pretty nicely.
>
> So I would argue, that Bruno's throughput values are very good taken the
> interference problem. But I'd say that the biggest problem of manets
> did not occur yet in Berlin, because most nodes are routers who don't
> issue packets on there own, but do only forwarding. So these values are
> pretty much because, there is not much congestion/concurrent traffic so
> far.
>
> @Bruno: Nice topology screenshot :-) I guess it was you who added the LQ
> feature. Very Nice. In case, you are still down with the topology
> plugin, I'd very much like to see some integration of the MIDs. I often
> find the routing table quite misleading, because there seem to be quite
> a few MIDs around. (maybe just having several IPs in one box would do).
> Anyway, I also like your view-topology script, and wonder if it could
> be changed to run only once, so that I could use it as a cron job. This
> way I could put the topol image on my web site. Maybe even a cgi option
> would be nice (I don't speak perl). Then you could have a look at the
> net from my point of view ;-)
> Also, if you feel like it you could do a "netperf -p 12865 -H
> 104.193.1.5". the x.x.x.5 is my IP and since we are both on the very
> west/east of the net, a test might be interesting. though, i don't know
> if the loops that currently in the net might harm this.
> Just my wishlist after xmas ;-) But I guess you welcome some feedback,
> and you know, just as time permits and you feel like it.
>
> good night,
> Jens
>
>