Re: [RFC v5 4/6] sched/fair: Tune task wake-up logic to pack small background tasks on fewer cores

[Dietmar Eggemann]

On 09/10/2019 19:02, Parth Shah wrote:
> 
> 
> On 10/9/19 7:56 PM, Dietmar Eggemann wrote:
>> On 09/10/2019 10:57, Parth Shah wrote:
>>
>> [...]
>>
>>>> On 07/10/2019 18:53, Parth Shah wrote:
>>>>>
>>>>>
>>>>> On 10/7/19 5:49 PM, Vincent Guittot wrote:
>>>>>> On Mon, 7 Oct 2019 at 10:31, Parth Shah <parth@xxxxxxxxxxxxx> wrote:

[...]

> ok. so does that mean TurboSched can still do some good in such systems as
> well ?
> I mean save energy even when rd->overutilized==1 by not waking user
> classified bg tasks on idle core.

I wouldn't say it is impossible but how likely would it be?

The Android runtime already classifies its tasks into groups such as
background, foreground, top-app, etc. It uses existing infrastructure
like cpusets, taskgroups, util_clamp (or its out-of-tree predecessor
schedtune) as well as EAS/Energy Model on asymmetric CPU capacity
systems to map them (differently) onto the CPU topology to achieve the
best possible performance/energy consumption trade-off.

Moreover, Google and Arm are keen getting the concept of 'latency nice'
upstream so we can map Android Common Kernel's 'prefer idle' feature
into the mainline energy-aware wu path.

So I'm afraid the question whether TurboSched could make sense on an
Android system can only be answered by people responsible for future
Android runtime architecture.