[Olsr-dev] LinkQualityMult problems in 0.5.6-r5

[Henning Rogge]

Am Samstag 05 September 2009 13:29:27 schrieb Mitar:
> I agree. I think current semantics of LinkQualityMult is problematic
> as it changes link quality only in one direction. I think it would be
> better if you (as an node owner/administrator) would be able to say
> that you would prefer that ETX of (all or some) links to your node
> would be (artificially) increased not by changing link quality (as it
> is still good to know what is real link quality).
The next generation of OLSR (the one with OLSRv2 packets) will only transmit 
"linkcosts" with TCs, so lq_mult will become some kind of linkcost_mult with a 
limited range.

"LQ" is just an internal from the ETX routing metric.

of course there might be (not decided at the moment) two cost values to handle 
asymmetric links (good in one, bad in the other direction). But this will need 
some changes in the SPF core.
 
> Or maybe oven better with ETX so that it would properly represent link
> quality but to add another multiplier of "willingness" to be a route
> (not to be confused with current OLSR willingness to relay control
> data). So that OLSR would choose route based on ETX * "willingness".
> 
> This would then solve the problem of administrating diverse networks.
> Like we could set this "willingness" on a VPN server to low and it
> would not be necessary to play with ETX. And some other node which has
> expensive to use interface would set "willingness" on that interface
> to low to discourage setting route through it.
Setting a global "willingness" for routing traffic for a node would be 
(mostly) a sane thing. You do it because you know something about your box, 
not because you fight with some neighbors. ;)

> > Backup link is just a name for "don't use it unless it's the only way".
> > Anything else should be handled by the routing metric.
> 
> And this "only way" mean 10% packet drop? 50%? 100%?
We will see, it's just an idea at the moment.
 
> > MANETs (mobile ad hoc networks, Freifunk is a special case of this) do
> > work because they use a sane routing metric on all nodes... and they use
> > THE SAME routing metric. You cannot really change global behavior with
> > local changes without lot's of problems for the rest of the network.
> 
> But they also work because of all diverse people with diverse reason
> why they cooperate. And if somebody want to join a network but be just
> a relay in a case of emergency it should be possible to setup this for
> him. Currently only way is to set LinkQualityMult.
> 
> And it should be depended on if other nodes agree that his/her node is
> just emergency node or not.
> 
> MANETs should work on a diverse range of equipment, platforms and people.
Another idea Markus and me discussed was to propagate your "lq-mult" (or 
whatever it will become) in the TCs through the network, so it's possible to 
debug global problems. And there could be some layer-8 discussions when people 
start to make a good link bad, just because they like their route to node X 
the other way around the block. ;)

Henning
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.olsr.org/pipermail/olsr-dev/attachments/20090905/9c566ab2/attachment.sig>