[olsr-dev] MRP selection (all OLSRd versions)

[Thomas Lopatic]

Hey Sven-Ola,

> (Mail with pdf attachment obviously halted. PDF is here:
> http://styx.commando.de/sven-ola/mprcoverage.pdf )

Hmmm. Wait a second. .1 sees .2, .3, and .4, and says in its LQ HELLOs
that it sees these nodes. .5 sees .6, .7, and .8, and also says in its
LQ HELLOs that it sees these nodes.

So, as long as, for example, .1 sees .5's LQ HELLOs, it should know that
.5 sees nodes that it does not see itself, namely .6, .7, and .8. Which
in turn should make .1 select .5 as an MPR.

Analogously, if .5 sees .1's LQ HELLOs, it should know that .1 sees
nodes that it does not see itself, namely .2, .3, and .4. So, .5 should
also select .1 as an MPR.

This assumes that .2, .3, and .4 do not see .6, .7, and .8, which seems
to be the case in the illustration.

So, I am not sure whether forcing one MPR per interface would change
anything and solve this particular problem. If there's a node at any
interface that sees nodes that I do not see, I make this node an MPR for
me. Hmmm. That should do the job, shouldn't it?

So, my guess would be that we've discovered a bug in the olsrd
implementation here and not a design error. Or am I missing something here?

It's pretty weird that .1 and .5 see each other but do not mutually
select themselves as MPRs.

-Thomas