[OLSR-users] Something to look at

[Andrew Nott]

Amit Kucheria wrote:
> Hi Guys,
> 
> On Wed, 2004-10-13 at 22:54, Andreas Tønnesen wrote:
> 
>>Hi Jhon,
>>
>>That was quite an interesting update!
>>Are you using hysteresis? Have you tried to experiment with different 
>>hysteresis values? ...still this might be of little help as the link is 
>>probably "seen" until everybody tries to use it...
>>But what about setting the lower threshold for reception on the 
>>interface(iwconfig sens)? Has anybody here done any experimenting with 
>>this parameter?
> 
> 
> The best solution IMHO is route weighting based on SNR. This was
> discussed 1-2 weeks ago. But the biggest problem facing everybody on
> Linux seems to be that different wireless drivers do not implement all
> the iw* wireless tools extensions.
> 
> Even if SNR is not available, would it be possible to weigh a link based
> on its perceived flakiness? That is, the number of times a route is
> dropped in a unit time. Higher the number, lower its weight.
> 
> Wish I had more time to implement the SNR solution though.
> 
> Regards,
> Amit

I ran across an interesting potential solution called ETX (Estimated 
Transmission Count) while investigating mesh protocols a few months ago. 
The basic idea is to base link weighting on an estimate of the total 
number of packet transmissions required to negotiate the link. So for 
example a hop suffering 50% average packet loss is given a weighting of 2.

This should provide much of the benefit of SNR based route weighting, 
without requiring access to low level information from the wireless drivers.

I've also seen a few references to people working on implementing ETX in 
OLSR, but only second or third hand, and never with a name or website 
attached.

Paper describing ETX http://citeseer.ist.psu.edu/641816.html

Andrew Nott


>>This might not help - but its worth a shot :)
>>
>>- Andreas
>>John Gorkos wrote:
> 
> 
> <snip> 
>