[Olsr-users] Scaling of olsr with reduced TTL

[Aaron Kaplan]

well, dijkstra sort of depends on the whole graph being present, right?
As opposed to voices in Berlin I however do not believe that this is  
a bad thing.
If you calculate how much a 10 000 entry routing table is (in memory)  
then you find out that it is not so overly much. for IPv6 128 bits, a  
very rough estimate shows that src,dest,extrainfos (16 bytes x 3) for  
10k entries is merely 480kBytes. Nothing. The big issue is how to use  
scalable algorithms (for example prefix search vs. brute force search  
etc) AND these algorithms should also be implemented in a *readable*  
and efficient way.

And yes, I agree: in general the trick will be to calculate as little  
as necessary.


On Nov 9, 2007, at 12:05 AM, Jens Nachtigall wrote:

>> AFAIK there are a few ideas in our heads at the moment: Hannes
>> mentioned something with incremental (diffs) of TC messages. This
>> should drastically reduce the load.
>
> Yes, such things are very nice and put the "point of not scaling- 
> anymore" into
> farther regions. However, still you would have such a point. I was  
> wondering,
> if such a point at which the network is too large to scale would  
> not exist
> anymore, if just the TTL was not 255 but just something like 8. No  
> artifical
> network splitting necessary, just one wireless network, and every  
> few hops a
> gateway.
>
>> But what you wrote below sounds very much like HSLS aehh...  
>> "fisheye".
>> Or did I simply misunderstand you now?
>
> Somewhat like fisheye, just that the max TTL is permanently reduce,  
> say 8
> instead of 255. So a node does not know anything about nodes  
> further away
> than 8 hops. I was just wondering if this works with Djikstra, if  
> the node

is that good?
Usually if you want to segment like that, you use BGP.

> only knows its neighborhood and not the whole topology (assuming  
> the node
> only wants to talk to this neighboorhoud within which might be an  
> HNA)?
>
> Don't think this is brainfuck, throwing out thousands of nodes  
> would be
> nothing for telcos or a government, also not for freenetworks with  
> some
> funding.
>
hm, do you mean that freifunk is underfunded and that throwing out  
thousands of nodes from the routing table solves problems? I just did  
not really understand the point.

curious about your ideas... :)
a.