Aubrey Li wrote:
> [snip]
>>> That is the behavior in the original linux implementation.
>>> Actually there is a mechanism to auto-adjust the ticktime, but it has been
>>> removed to powertop prototype quickly delivered, it would be great if it
>>> can be
>>> recovered. Think about the future tickless kernel, the cpu wakeup times
>>> become
>>> less, making the ticktime longer makes more sense if the wakeup events is
>>> few
>>> in the current sampling period.
>> So you mean adapting the interval based on the amount of wakeups - i.e few
>> wakeups, higher interval and vice-versa ?
>>
>> I'll have a look at the original code.
>>
> That would be great, thanks, :-)
Ok, the Linux version sets maxsleep (below) to the duration of the last
transition and then sets the interval according to
if (maxsleep < 5.0)
ticktime = 10;
else if (maxsleep < 30.0)
ticktime = 15;
else if (maxsleep < 100.0)
ticktime = 20;
else if (maxsleep < 400.0)
ticktime = 30;
else
ticktime = 45;
There's also
if (wakeups_per_second < 0)
ticktime = 2;
How about adding this functionality as the default option, and fixing
the interval if -t is used ?
thanks
Rafael
