[olsr-dev] dropping OLSRd-pakets / VPN

Daniel Poelzleithner (spam-protected)
Thu Nov 9 20:34:40 CET 2006

Hash: SHA1

Bastian Bittorf wrote:

> Yes, but we need this in our special case.
> Long story short: 
> The dirty solution is now to drop about 90%
> of all wired olsr-packets with a 90% gain of olsr traffic.

Basti, once again your solutions are shocking :)

I experimented with the nth and random iptables modules and found got
some very strange results, that may explain the regular 0.0 ETX
phenomena we see (at least i see it often).
For example: I had a good link to the uplink node on the opposite street
side. Mostly ETX 1.3-2.5 (it's a Steet with lots of Nodes).
Then the ETX suddenly drops down to 0.0 because one side has LQ 0.0, but
just for 10-20 secounds. The link still works in that time. But the time
is long enough for olsr to switch to alternative routes and switch back
some secounds later. Very annoying when using ssh ;)

Playing with nth I experienced the problem very often:

Chain OUTPUT (policy ACCEPT 8 packets, 2646 bytes)
 pkts bytes target     prot opt in     out     source
15006   21M DROP       udp  --  *      *           udp dpt:698 every 3th
 7618   11M ACCEPT     udp  --  *      tap0           udp dpt:698

This simple rule, even when most of the packets go through, causes very
often NLQ's of 0.0 and the link is broken. It seems to depend on . I
tried experimenting with TTL, but all packets, regardless of fisheye
turned on or off, have TTL 64, making it hard to write good rules to
drop the right packages.
If the wrong packets are dropped, the link is rated much worser then it
actually is.

> Now we could need an LQ-mechanism with
> values greater then "1.0" to make obviously bad
> ETX's better.

This is alread working my using a Multiplier > 1.0 (at least in the the
freifunk version). I reduced the ETX of a lan connection to 0.1 for
example :)

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


More information about the Olsr-dev mailing list