[olsr-dev] LQ code enhancements

[Sven-Ola Tuecke]

Hi elektra,

in practice, I observe diversity in OLSR packet sizes. OLSR Packets between 
28 Bytes and 1400 Bytes UDP payload can be seen on my WRT devices. Another 
observation is, that on relatively bad links the packet loss is correlated 
with the packet size. Which means: You will get ping answers but cannot 
contact the same web site until you set MTU=256 or similar.

But I fully agree to the "do the homework first until going further" thing 
Thomas expressed earlier. For my situation here (with fixed stations) manual 
tweaking with the LQ multiplier option and/or with the brute force iptables 
block-that-link method will do. Note: The LQ/ETX works great, but manual 
tweaking may work better ;-)

Regards,
Sven-Ola

"onelektra" <(spam-protected)> schrieb im Newsbeitrag 
news:(spam-protected)
> Hi list -
>
> the idea of weighting hello-paket-sizes came up with the introduction of 
> the LQ-extension. I was discussing this with Thomas by the time he 
> implemented the LQ-mechanism.
>
> I suggest that the core of the routing-deamon has to proof that it is 
> stable before considering such finetuning.
>
> It is a point to look at. But is it worth the effort? I m not quite shure. 
> Why weight paketsizes? The paketsize of hello-messages increases if a node 
> has more neighbours. Since we still use only one interface having more 
> neighbours also means having more collisions.
>
> So we are not weighting the paketsizes, but in practise a node with many 
> neighbours faces more paketloss of hello-messages not only cause it has to 
> send bigger hello-messages. Considering the node has to send bigger pakets 
> will be more accurate but palliate its situation.
>
> I think the effect of introducing this into the code is marginal. It 
> certainly increases the complexity of the program and consumes more 
> CPU-power.
>
> cu elektra
>
>
>> Hi Sven-Ola,
>>
>>> just an Idea, I would like to mention. Is it possible to consider the 
>>> OLSR
>>> packet size when calculating LQ/NLQ? A big 1400 byte packet should have 
>>> more
>>> weight than a short 28 byte mini packet. As far as I can tell, the LQ 
>>> code
>>> just counts packets...
>>
>>
>> Thanks for sharing your thoughts. I also think that this is something to 
>> consider in the future. I think that many interesting ideas have been 
>> voiced on the mailing lists concerning an improved, configurable, and 
>> flexible metric that is a superset of ETX, and that weighting small 
>> packets different from larger packets is something that we should look 
>> into.
>>
>> I fully agree that we should experiment with these things, but first I'd 
>> like to make the current link quality code stable, so that we have a good 
>> foundation from which to start.
>>
>> -Thomas
>>
>>
>> _______________________________________________
>> olsr-dev mailing list
>> (spam-protected)
>> https://www.olsr.org/mailman/listinfo/olsr-dev
>>
>>
>
> _______________________________________________
> olsr-dev mailing list
> (spam-protected)
> https://www.olsr.org/mailman/listinfo/olsr-dev