cvs commit: src/sys/i386/cpufreq est.c

[Robert Watson]

On Mon, 17 Mar 2008, Poul-Henning Kamp wrote:

> In message <200803170933.48212.jhb at freebsd.org>, John Baldwin writes:
>
>> Hmm, I actually consider this a feature when I'm not running powerd to use 
>> less battery.  I think we should only bump up the CPU on battery power when 
>> using powerd so that it can be lowered again to save battery power when the 
>> CPU is idle.
>
> We have cpufreq enabled by default now, badly configured machines run at 50% 
> of rated CPU power because people don't know that they need to enable 
> powerd(8) on servers.
>
> This is only going to get worse when more EnergyStar compliant servers hit 
> the channel.
>
> I think setting full speed is the correct choice, if people care about 
> powersaving, they need to configured it, if they don't they should get their 
> moneys worth out of their hardware.

If cpufreq is going to be enabled by default, should we be enabling powerd by 
default, or at least having a powerd_enable="AUTO" that detects the 
appropriate frobs and feedback sources and turns on powerd if it's going to be 
useful?  There might be a reasonable argument to be made that in two of the 
three computing environments of choice for FreeBSD (notebooks, servers in 
colos), power management is a basic assumption and we should turn on the 
necessary bits to deal with it.

Robert N M Watson
Computer Laboratory
University of Cambridge