Re: [RFC PATCH 0/7] Introduce thermal pressure

[Lukasz Luba]

Hi Vincent,

On 10/16/2018 07:11 PM, Vincent Guittot wrote:
> Hi Lukasz,
> 
> On Thu, 11 Oct 2018 at 13:10, Lukasz Luba <l.luba@xxxxxxxxxxxxxxxxxxx> wrote:
>>
>>
>>
>> On 10/10/2018 07:30 PM, Thara Gopinath wrote:
>>> Hello Lukasz,
>>>
>>> On 10/10/2018 11:35 AM, Lukasz Luba wrote:
>>>> Hi Thara,
>>>>
>>>> I have run it on Exynos5433 mainline.
>>>> When it is enabled with step_wise thermal governor,
>>>> some of my tests are showing ~30-50% regression (i.e. hackbench),
>>>> dhrystone ~10%.
>>>
>>> That is interesting. If I understand correctly, dhrystone spawns 16
>>> threads or so and floods the system. In "theory", such a test should not
>>> see any performance improvement and degradation. What is the thermal
>>> activity like in your system? I will try running one of these tests on
>>> hikey960.
>> I use this dhrystone implementation:
>> https://github.com/Keith-S-Thompson/dhrystone/blob/master/v2.2/dry.c
>> It does not span new threads/processes and I pinned it to a single cpu.
>>
>> My thermal setup is probably different than yours.
>> You have (on hikey960) probably 1 sensor for whole SoC and one thermal
>> zone (if it is this mainline file:
>> arch/arm64/boot/dts/hisilicon/hi3660.dtsi).
>> This thermal zone has two cooling devices - two clusters with dvfs.
>> Your temperature signal read out from that sensor is probably much
>> smoother. When you have sensor inside cluster, the rising factor
>> can be even 20deg/s (for big cores).
>> In my case, there are 4 thermal zones, each cluster has it's private
>> sensor and thermal zone. There is no 'SoC sensor' or 'PCB sensor',
>> which is recommended for IPA.
>>>>
>>>> Could you tell me which thermal governor was used in your case?
>>>> Please also share the name of that benchmark, i will give it a try.
>>>> Is it single threaded compute-intensive?
>>>
>>> Step-wise governor.
>>> I use aobench which is part of phoronix-test-suite.
>>>
>>> Regards
>>> Thara
>>>
>> I have built this aobench and run it pinned to single big cpu:
>> time taskset -c 4 ./aobench
> 
> Why have you pinned the test only on CPU4 ?
> Goal of thermal pressure is to inform the scheduler of reduced compute
> capacity and help the scheduler to take better decision in tasks
> placement.
> So I would not expect perf impact on your test as the bench will stay
> pinned on the cpu4
> That being said you obviously have perf impact as shown below in your results
> And other tasks on the system are not pinned and might come and
> disturb your bench
Unpinned runs on this platform have a lot of variation. The tests take
even 16min to finish.

Regards,
Lukasz
> 
>> The results:
>> 3min-5:30min [mainline]
>> 5:15min-5:50min [+patchset]
>>
>> The idea is definitely worth to investigate further.
> 
> Yes I agree
> 
> Vincent
>>
>> Regards,
>> Lukasz
>>
>>
>>
> 
>