[Olsr-users] OLSRd Scailability

[Benjamin Henrion]

On Thu, Jul 31, 2008 at 12:10 PM, aaron <(spam-protected)> wrote:
> Henning Rogge wrote:
>> Am Donnerstag 31 Juli 2008 11:34:05 schrieb Derek C:
>>
>>> Hi all,
>>>
>>> I'm sure this has been discussed a lot before but I can't find out
>>> anything exact on the net.
>>>
>>> Is there a scalability "wall" with OLSRd?
>>>
>
> Well, I must add: usually other mesh routing protocols start to have a
> "wall"
> at ca. 100 nodes :))
>
>
> So, in a sense - OLSR seems to me as still one of the most scalable
> *tested* mesh protocols that I know of
> (maybe excluding hierarchical protocols). Please send links to more
> scalable protocols in case you know of them!
> Maybe we can learn from them :)
> Pure papers don't count (except for inspiration). Since scalability
> needs to be tested in practice.
>
>> Yes... at the moment you run against a wall because OLSR i
>> s a little
>> bit "chaty" (it creates too many packages for routing). It will happen only
>> for larger networks (500+ nodes), 1000 nodes seem to be possible even with
>>
> How many does freifunk have? Probably around 1000 by now.
> How many does the athens wireless network have ?
> Around 2300 (BGP + OLSR). c.f.: http://wind.awmn.net/?page=nodes
>
> So, I am pretty confident that we well passed the 1000 nodes wall.
> The next would be 10000 nodes. And I agree with Henning - once we reduce
> the number of packets in the air
> we will be able to scale higher. Dijkstra calculations don't seem to be
> the problem anymore (thanks to the OLSR-NG efforts).
>
> So to sum it up - I guess OLSR is already a pretty mature thing for
> starting a new network.
> However there is - as Henning suggested - *ample* room for improvement.
> Especially in code readability, metric calculation, reducing the number
> of packets in the air etc etc.

I am pretty sure tuning the config file would already achieve a lot in
terms of reducing airtime.

If you have a static network with no mobile nodes, it is acceptable to
reduce the routing information to one packet per minute. Or even 5
minutes would be acceptable. Or 10 or 15 minutes. Non-mobile nodes
don't need to have a frequent update of their routing table.

I am going to Berlin this week-end until the 6, and I am gonna work on
a graphical interface for the UML-wifi QT interface, see here:

http://folk.uio.no/paalee/referencing_publications/ref-nr-guffens-presentation05.pdf

--
Benjamin Henrion <bhenrion at ffii.org>
FFII Brussels - +32-484-566109 - +32-2-4148403