[olsr-dev] Fixing the olsr routing mistake

[Daniel Poelzleithner]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi,

we discuessed this matter deeply on the wlanware mailinglist some time
ago and I will implement a patch soon.

Here the problem:

Node A  <----------2.5---------> Node B
        \--1.5-- Node C --1.5--/

The Problem is, how olsr currently calculates best routes by summing up
all etx values of all connections in the path. The problem is, that
every hop has a penalty of 1.0 by the way etx is calculated. Tests in
our fairly large mesh network shows, that throughput increases if Node C
is used instead of of the direct link A<->B. The probability of a lost
packet in A<->B is 1.5 while A-C-B only has 1.0. Olsr choses the direct
connection, because of the penalty every hop receives.
The problem is even more visible when LAN connections is used on some
connections, then the packets are often routed not through lan, which
would be much faster and better.

My solution is following.
Instead of simply sum up the etx values do:

sum(etx's) - n*X

n is the nummer of connections and X a float value. 1.0 would mean no
hop penelty at all and 0 is the current way of doing it. This would
allow to control if more hops or more direct connections should be used.

I'm not into the olsrd code, so my question is: Is there a way to
implement this, without causing routing loops if X is not the same on
all Nodes in the network ?
I imaging that, if a node doesn't have the patch or X is different,
constant routing loops will happen.

kindly regards
  Daniel
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: GnuPT 2.7.2
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFFUf7sy/mkIQp7AD0RAh8YAKDZzG3ygogDGa1J9FDgQ5cw5pDQTACgyxee
Ae2FTkiPHPNa9N7hgqg/S98=
=zzgC
-----END PGP SIGNATURE-----