Automated adoption of devkit-power latencies

[Harald Hoyer]

Phil Knirsch wrote:
> Richard Hughes wrote:
>>>>> I'd personally like to use devkit-power and it's API as a service, as 
>>>>> overloading it with such functionality might very quickly lead to the 
>>>>> problems we've seen in the past with HAL where everything was put into
>>>>> a single service.
>>>> No, I think it's 100% on topic for DK-p. Powersaving is inextricably
>>>> linked to latency in my opinion.
>>>>
>>> What about cases where you want to completely shut down a device though 
>>> as you know it's not used?
>> Well, I think for the sort of latencies we are talking about (us and low
>> order ms) we need to do this in the kernel driver by default. The logic
>> goes like this: If the device isn't being used, power it down _UNLESS_
>> it takes more time than $CURRENT_LATENCY_SETTING to wakeup.
>>
>> This way everything sleeps by default.
>>
> 
> Yep, absolutely agree. But even there wouldn't it be nice if e.g. during 
> the day that server would have a 10us latency for CPU and during the 
> night automatically transition to higher and higher latencies, up to 
> even 100ms or so? Or a desktop system that has a screenlock running 
> overnight, the user forgot to close his firefox window with a flash 
> plugin running in it. Thats what i'm thinking about at least. Not to 
> have a fixed latency but a dynamic one that, depending on whats going on 
> on the system adapts those.
> 
> And devices that take a lot longer to power down resp. up e.g. harddisks 
> aren't addresses by the latency changes yet if i remember correctly?
> 
> Regards, Phil
> 

How about a totally dynamic latency, which grows as soon as network traffic per 
hour decreases and the other way round.