[Olsr-dev] feature request for on demand interface discovery

[Henning Rogge]

Olsrd is already monitoring interfaces via netlink...

unfortunately (in some cornercases) this fails, thats why the timer is there.

Can you just set NicChgsPollInt to 60 seconds as an experiment, to see
if it solves your CPUload problem?

Henning

On Tue, May 6, 2014 at 1:55 PM, cstorch <(spam-protected)> wrote:
> hi there,
>
> we have a pretty special setup here at Freifunk Weimar - "Weimarnetz".
> Our root server is our vpn gateway for internet providing dsl gateways.
>
> There is olsr version
> "(spam-protected):~$ olsrd -v
>  *** olsr.org -
> 0.6.6.1-git_0000000-hash_6fed9f54908463039e59671dc6de44c8 ***
>  Build date: 2013-12-05 15:48:04 on weimar2
>  http://www.olsr.org"
> running.
>
> To keep our vpn setup easy we configured 1000 tap interfaces, one for
> each possible vpn client. The olsr config contains also 1000 taps.
>
> When olsr is startet, the load of the server goes up to "2". (it is a vm
> on multicore xeon with enough memory)
>
> We did some analysis with strace:
> 4:24:56.186377 ioctl(3, SIOCGIFFLAGS, {ifr_name="tap635", ???}) = -1
> ENODEV (No such device)
> this is repeated for all interfaces and takes 12seconds to complete.
> After (NicChgsPollInt) the interface discovery is repeated.
>
> Our idea is to switch off the "NicChgsPollInt" and use some kind of
> "kill -USR1  pidof olsrd" (like for dd) to trigger a interface discovery.
>
> What is your opinion?
>
> cstorch
>
> --
> Olsr-dev mailing list
> (spam-protected)
> https://lists.olsr.org/mailman/listinfo/olsr-dev