[Olsr-dev] Little change to olsr core

[Teco Boot]

Op 27 jul. 2012, om 13:07 heeft Henning Rogge het volgende geschreven:

> On 07/27/2012 01:01 PM, Teco Boot wrote:
>> Op 27 jul. 2012, om 11:28 heeft Henning Rogge het volgende geschreven:
>>> The current core doesn't support asymmetric links.
>> Could be.
>> When all nodes send out their own TC (no topology reduction), all
>> with outbound costs, what is the problem?
> 
> At the moment there is no such thing as "outbound/inbound" cost in our TCs. LQ/NLQ are always combined into a symmetric value.

Now, we disabled topology reduction, so all nodes send out TC (to be checked).
The costs can get semantics that it is for outbound. 

//

>>> Setting your local ETX to "a" but having your neighbor announcing an
>>> ETX of "b" sounds like a lot of trouble.
>> But this is normal during convergence. Local ETX is changing more
>> rapidly than arrived messages, send out with fixed intervals.
>> (yes, there could be loops)
> 
> Might be interesting to only use the LQs from the sent TCs instead of the local measurements to decrease the chance of loops.
Yes.
(take TC cost, not LQ)

//

>>> Whats about this, instead of implementing a static LQ/ETX, we could
>>> allow a user to define a maximum LQ for a link (between 0 and 1).
>>> Calculation will be running as normal, just that it doesn't go higher
>>> than a certain value.
>> We can make a new lq_plugin, for experimentation. Based on
>> lq_plugin_default_ffeth. All enhancements are configured, so default
>> is 100% compatible. Messages should be compatible too, for mixed mode
>> operation. Candidate enhancements:
>>  - hysteresis (suppress frequent small changes), with config option
>>    (LQHYST 0.00 - 0.99)
> 
> The hysteresis is already there, just not with tuning parameters.

Can you point me to the code? I can't find it.

(hysteresis != dampening)

//

>>  - LQ window size
>>  - max costs and max LQ thresholds
>>  - mixing available L2 feedback with LQ, such as RSSI / DLEP info.
> 
> I think we should push this to a different experiment.

Test one by one: yes.
Postpone DLEP: I guess you are right.

Merging RSSI with loss ratio is not that difficult. 
Problem here: OLSR has the loss ratio, DLEP has the L2 feedback.
So maybe just do some experiments with RSSI.

Because of required config to make it active, we could have a shared experimental plugin.

> 
>>  - a more exponential increase of costs, so links with higher losses
>>    have much higher costs could be something like
>>    ( 255 * 255 * 255 ) / ( LQ * NLQ * NLQ)
> 
> Not easy because of the small range of possible values for LQ/NLQ.

My proposal is a small change in cost calculation. I expect high benefits. ETX does a better job than hop-count, but what I see is that links with cost just below ETX=2 don't perform as well as a two-hop path with ETX=1.


Teco