[Olsr-dev] Seeking comments: OLSR+ETX v/s DSR+ETX

[elektra]

Hi -

just a quick note to this:

Before we decided to choose ETX as metric for olsrd we were experimenting with 
measuring round trip times. Axel Neumann tried to implement this while we 
were still using mobilemesh, but the measurement results seemed to be quite 
arbitrary. But I assume this was a problem of our implementation, because any 
implementation of ping does a much better job than what we had achieved.

And btw. I don't see any irrational 'voodoo' in LQ/NLQ/ETX - at night I can do 
downloads in our mesh via 18 hops with approx. 12 kByte/sec... Some people 
claimed that a mesh only works up to three hops. So the 'voodoo' seems to 
help a lot ;-)

Back then in 2004 I just took the idea of ETX as a base and modified it - 
instead of sending bursts of extra packets we do statistics about Hello's, 
and we are telling our neighbors with every Hello about the LQ values we are 
measuring. I think that is/was very nifty because we don't have to send extra 
packets like it was initially suggested in the ETX  paper.

So I still think it would be a valuable idea to measure RTT for different 
packet sizes. And the payload of these packages should transport something 
useful... For example the results of the last RTT-measurement could be 
included. 

On the other hand I wonder how useful the overhead is in practise.

cu elektra




> Sven-Ola Tuecke wrote:
> > For ETT we need a round trip time to calculate. To measure/evaluate the
> > RTT difference between large and small packets. So basically, you will
> > send out a packet and listen for the answer, checking the time.
>
> the response part does not reflect my understanding. -
>
> what i am envisioning is:
>
> 1. a unicast small packet plus a unicast large packet.
> 2. both packets shall carry timestamps and be sequence
>     number protected.
> 3. based on the arrival time the receiver can calculate the
>     air-bandwidth and
> 4. attach that rx-bandwidth to a neighbor advertisement
>     in a new TBD OLSRDv2 TLV
> 5. during SPF calculation the average bandwidth between
>     a pair of neighbors is computed. the inverse
>     average bandwidth is multiplied with ETX.
>
> /hannes