On Tue, Oct 16, 2007 at 09:40:03AM -0700, Stephen Hemminger wrote: > On Mon, 15 Oct 2007 23:48:12 +0200 > Michał Mirosław <mirq-linux@xxxxxxxxxxxx> wrote: > > Quick CPU eater kill. This might be not the best fix possible. > > Without this patch, hello_timer is firing 250 times/sec and eating about > > 8% CPU time on my an otherwise idle system (dynticks enabled). 2.6.21.5 was > > even worse - 20% (HZ=1000). (br->hello_time was supposedly 0 when STP was > > disabled). > Good idea, but patch needs work. > > The default hello_time is 2 seconds, so if it is going up to 250/sec then > maybe other root bridge is updating it. > I don't see where hello_time gets set to 0 if STP disabled? Huh! I now checked my config and it turned out, that I set hello_time=0 there (maybe to disable hellos...) and forgot about it. It's removed now - just in case. > My preference would be to have hello_time set to zero if not running STP > and not start timer if it is zero. What about: brctr stp br0 off brctr stp br0 on This sequence would disable HELLO's unless you would keep backup of this field and restore it while turning STP back on. > Also, the check for not sending config BPDU is separate. We do was time > deciding to send BPDU when STP is disabled, the existing test could > just be moved up the call chain. That would move the STP logic to br_device.c. It's mixed up anyway, but maybe I'm missing some point - I expected that by disabling STP I'm turning my Linux box to a dump switch, that doesn't even care about other switches nor forwarding loops that may happen (and so doesn't need to send nor receive any BPDUs). -- Michal Miroslaw _______________________________________________ Bridge mailing list Bridge@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linux-foundation.org/mailman/listinfo/bridge
