[Debian-eeepc-devel] linux kernel, CONFIG_HZ_ and netbook

Trent W. Buck twb at cybersource.com.au
Tue Apr 6 09:55:44 UTC 2010


Ben Armstrong <synrg at sanctuary.nslug.ns.ca> writes:

> On 25/02/10 06:29 AM, Ben Armstrong wrote:
>> On 25/02/10 06:22 AM, giggzounet wrote:
>>> Hi,
>>>
>>> I'm compiling the last kernel. And I have a doubt on the choice of
>>> CONFIG_HZ : 250, 300 or 1000Hz ? On desktop computers I always set
>>> 1000Hz, but for my netbook I don't know : does the computer consume
>>> more energy with 1000hz ? Do you have Papers on this topic ?
>>
>> I don't know.  Outside of the scope of our project.
>>
>> I assume Debian has chosen reasonable defaults:
>>
>> CONFIG_NO_HZ=y
>> CONFIG_HZ_250=y
>> CONFIG_HZ=250
>>
>> The first one makes the stock Debian kernel 'tickless' which results
>> in considerable power savings:
>>
>> http://www.phoronix.com/scan.php?page=article&item=651&num=1
>
> Huh.  [...] They said there was hardly any savings.

Using Debian's .33 kernel, "powertop -dt15" reports a wakeups-from-idle
of 24Hz.  I assume with CONFIG_NO_HZ=n and CONFIG_HZ=250 you'd get a
minimum of 250Hz -- meaning less time in C4 and more time in C0.

While I haven't performed any tests with CONFIG_NO_HZ=n, I would
attribute Larabel's findings to:

  1) he tested a server-oriented CPU (Clovertown).  AFAICT from
     Wikipedia, it doesn't implement CPU states above C1, which is where
     the power savings ought to really kick in.

  2) his definition of "idle" might mean a full desktop environment that
     he isn't actively interacting with -- i.e. lots of daemons.  Even a
     simple *idling* python daemon (http://paste.debian.net/67627/) is
     enough to drive my "idle" rate from 24Hz to around 100Hz.




More information about the Debian-eeepc-devel mailing list