On Mon, Jan 09, 2006 at 10:10:37AM +0100, Frederick Page wrote:
> Hi everybody,
> I had some issues with freezes, both on the WRAP and Soekris net4801.
> Earlier posts from this ML suggested to check temperatures and I tried
> operating the devices with open casing (which seemed to help). However
> it was only yesterday that I pressed my index finger onto the CPU chip
> and tried to hold on for about 15 seconds.
> I could not. Tested with both devices: both CPUs got hotter than I
> could bear. Then I remembered this in my XML file:
> <shellcmd>sysctl kern.polling.enable=1</shellcmd>
> which gave me astonishing low CPU usage of about 2%, even with my 2
> mBit SDSL fully saturated. To make a long story short: I removed that
> setting, restored the config and now have a noticeably cooler CPU,
> although CPU usage is at about 14 - 20% on saturated bandwidth.
> I have no explanation for this phenomenom, but wanted to share the
Well, it's hardly a surprise, is it? With device polling, your CPU
spends a whole lot more time in busy loops and much less time in the
idle state (I presume that like any other O/S, FreeBSD puts the CPU in
HLT state when it has nothing to do), hence the increased temperature.
Somebody correct me if I'm wrong.
Now IMO proper hardware should be stable even with CPU in a busy loop.
Why your CPU seems to get unreasonably hot in such a case, causing
stability issues, is beyond me.