[Olsr-dev] OLSRd 0.6.0 with ETT available

[Erik Tromp]

> I'll re-evaluate this.
> 
> We updated old version a bit:
>  - add more bandwidth values
>  - added hysteresis
>  - speed up convergence
> These made the plugin more useful.
> I'll get updated this version and send patches later on.

That would be nice.


> I have some ideas to improve the concept. To list a few:
>  - remove distinct data rate values completely

This already happens to some extent in the olsr-lc link_probe
plugin; see the file lib/link_probe/src/Probe.c lines 297 and
onward. The extremes are taken out of measured set.


>  - use first measurements instead of (possibly very wrong) default

Good idea. However, it may also be better to act like the
saying: reputation arrives on foot but departs on horseback
(perhaps a bad translation, I'm sorry :-)). In other words, a newly
discovered link should be well tested before it can be trusted
to route our precious traffic onto. It may be better not to
switch directly toward a new link that could turn out to be very
unreliable; that will create a lot of (topology) updates
in the network. By the way, it seems that the code, which
prevents topology and other updates in the case of small
changes in the link cost, has been removed from olsr version
0.6.0, compared to previous versions. Is that true?


>  - auto-calibrate RSSI. This implies passive probing.

This may be difficult. See
http://pdos.csail.mit.edu/~rtm/papers/p442-aguayo.pdf
(for example Figure 14). The relation between RSSI and
bandwidth may be hard to estimate.


>  - combine measured cost with ETX
> 
> Can we transform this to a lq-lc plugin? 

This is already the case. See the function calc_link_cost_ett(...)
in src/lq_plugin_ett.c , line 147.


> Then, add this in the repository?

Yes, that would be a good idea. I had to make quite some changes to
'standard' olsr-0.6.0, though, to make it suitable for lc routing.

Regards, Erik