Kernel panic when unpluggin AC adaptor
oberman at es.net
Thu May 13 16:39:04 UTC 2010
> Date: Thu, 13 May 2010 00:09:05 -0700
> From: perryh at pluto.rain.com
> "Kevin Oberman" <oberman at es.net> wrote:
> > ... TCC and throttling ...
> > they were intended for thermal management, not power management.
> Shouldn't the two be equivalent? Heat generated is directly related
> to power consumed.
They are roughly equivalent. TCC performs slightly better as it is
internal to the CPU and can run automatically with no OS involvement.
Throttling requires the OS set the number of cycles to skip on pins of
the CPU. My testing has shown only about a 5% difference on the worst
case in a P4-Mobile CPU. FreeBSD will use TCC when it is available and
throttling when it is not. It never uses both. (It did for a short while
and it was not pretty!)
Heat generated is almost directly proportional to power consumed.
Unfortunately, when you use TCC or throttling the performance is also
directly proportional to the power consumed. the result is that, to
perform any operation, you end up using almost the same power. Testing
has shown that it is really a net loss to do this.
For an excellent review of the situation, please read Alexandar Motin's
paper on FreeBSD power tuning at:
He performed the same testing I had performed about two years prior and
went well beyond what I had done. More importantly, he wrote up an
excellent document which describes how to set up a system for best
power/performance and explains why. I think is is a 'must read' for
anyone who wants to run a power efficient system.
R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: oberman at es.net Phone: +1 510 486-8634
Key fingerprint:059B 2DDF 031C 9BA3 14A4 EADA 927D EBB3 987B 3751
More information about the freebsd-stable