[OLSR-users] OpenWRT experimental and OLSR
Fri Apr 1 19:56:27 CEST 2005
Well.... the problem is that I run it on my core takeout router (I have two
gateways, but this one is the most heavily loaded), and the way I know that
it dies is when customers start calling me to complain that their internet
service is down. So I'm not entirely inclined to do a lot of testing with
it. However, I _will_ run the latest CVS code (checked out yesterday) in a
non-stripped mode in an effort to find the culprit.
I've discovered that, after spending hundreds of hours writing scripts to
constantly monitor the status of my wireless network, that the customers are
the best early warning system I could ask for. Plus, when I can immediately
fix what's broken, I gain the aura of a computer uber-god who can make the
internet appear at will. It tends to counter-act the "Why am I paying good
money for crappy service" questions...
Stay tuned for updates on the latest cvs.4.9 build performance in the
real-world. BTW, I don't use LQ, just hysterisis.
On Friday 01 April 2005 11:10, Thomas Lopatic wrote:
> Hi John,
> > I had some stability issues with the last CVS build I made, the daemon
> > would sometimes die silently.
> Would it be possible to run an unstripped executable and make it dump
> core? I guess that doing something like "ulimit -c 10240 ; olsrd" will
> do the job. It would be great if you could then send me you unstripped
> executable and the core dump.
> I'd love to see where the crash happens. We now know that 0.4.8 contains
> quite some show stoppers in the link quality code. 0.4.9 fixes those
> that are known to us. However, you seem to have found a new one. Time to
> investigate this for 0.4.10.
> Oh, BTW, when exactly did you check out the source code from CVS?
> > Finally, I've noticed that the load average on my core router is down
> > significantly with the new executable. Before, it would run ay .40-.55,
> > now it barely registers. Was there an errant loop that was sucking clock
> > cycles before?
> No, just embarrassingly bad algorithms. :-)
> olsr-users mailing list
More information about the Olsr-users