Analyze and optimize Android apps power consumption

Intel® Energy Profiler power analysis for Android* Alexander Weggerle

Copyright © 2014, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. Optimization Notice

CPU C-States / P-States

Intel Confidential 2

C1

C2

C3

C4

C5

C6

Pn

P1

P0 CPU

Active

CPU

Sleep

P0 - CPU active at highest frequency (HFM)

Pn - CPU active at lowest frequency (LFM)

C0 - CPU active (In any P-state)

C0

C1 - Core clock is Off

C3/C4 - Reduced Voltage, Partial L2 cache flush

C6 - Core Off, L2 cache flush, state saved to SRAM

The deeper the sleep state

more power saving

but longer to wake up

Po

we

r H

igh

er

La

ten

cy

Gre

ate

r


CPU Sleep States

• Flexible C-States to Select Idle Power Level vs. Responsiveness

Core voltage*

Core clock

PLL

L1 caches

L2 cache

Wakeup time*

Idle power*

C0 C1 C3 C4

off

Active state

off

flushed

off

off

flushed

off

partial flush

active

* Rough approximation

C6

off

off

off

off


Processor Power and Processor Frequency

9

59

109

159

209

259

309

359

0 0,5 1 1,5 2 2,5 3 3,5

Po

we

r (w

)

Frequency (GHz)

Power vs. Frequency Curve for Single Architecture

Copyright © 2004 Intel Corporation. All rights reserved. 5

Small Increases in Processor Speed

Results in Large Increases in Power


Components of power analysis Idle vs. Active

Do nothing efficiently

Hurry up and get idle.

e.g. Multi-threading (distributing work evenly across cores)

Optimize Sleep Behavior

Minimize sporadic wakeups.

Schedule all periodic activities from the app into same wakeup period.

What is waking h/w from low power states? Why?

Optimize Utilization

What is active? Why is it active?

Minimize Polling Loops. Use event driven framework when possible.

Turn devices off. Open devices can prevent the system from entering power saving state.


Why Minimize Wakeups?

Frequency of software activity makes big difference

Long uninterrupted

Idle period, is the key

7

4

5

6

7

8

9

10

11

12

0% 2% 4% 6% 8% 10%

Pla

tfo

rm P

ow

er

CPU Utilization

Solid Activity

Fragmented Activity

Co

nso

lid

ate


Types of Analysis

Two flavors of SoCWatch use cases:

Expert analysis of results

Comparison of results by non-experts

Focus on area of expertise: • A specific SoC device • A specific driver module • A specific application • etc…

Look for differences: • Wakeup frequency • P/C state changes • etc…

Best to have a baseline for comparison in all cases

Power Analysis with VTune™ Amplifier for Systems

9


Summary View

6/16/2014 11


C/P States Tab

6/16/2014 12


C - states

6/16/2014 13


Analyze Wakeups

Intel Confidential 14

Find dominant wakeups


What C-State residency numbers tell you

Residency in the deepest C-state should be >95% @ idle

If you see high residency in C0 state

CPU is executing instruction

Next Step: Active Analysis

If you see high residency in the intermediate C-states

Frequent active/idle transition is keeping CPU from entering deepest C-state. Possible causes are:

Application scheduling periodic timer with short period activity Application waits for interrupts (from device, IPI) very frequently

15


CPU Power Analysis Use case: Collecting data from sensor

• Continuously collecting accelerometer data • Use power analysis to find out impact of sampling frequency

SENSOR_DELAY_FASTEST SENSOR_DELAY_NORMAL

Lowest Power consumption

High Power consumption


Component Device States Find Components Wasting Power

17

Intel Device States:

DOi0 = On

DOi1-DOi2 = Intermediate

D0i3 = Off

Find:

• Which Devices are on/off?

• For this MP3 Playing video encoder is not used?

• When they got turned on/off?

• When a device is not in use… the software needs to turn it off


Correlate Metrics to find patterns

19

See all the metrics at once

Find patterns

Find the cause


Depends on which SoC

C-States

P-States (CPU frequency)

SOi States (System State)

Android Wakelocks

Temperature (Core, Skin, SoC, PMIC)

S(outh) C(luster) Device States

N(orth) C(luster) Device States

20

Potential Collections

Power Analysis Use Cases

21


Use cases: Idle

Comparison of system running at Idle with a system running at idle without the ability to go to sleep.

6/16/2014 22


Use cases: Idle

6/16/2014 23

Percentage of time in C6

Core 1: 62.7%

Core 0: 63.3%

Reasons for waking up

Kernel Timer: 4347

User Timer: 4120

IRQ 46: 570

IRQ 93: 557

Frequency is almost 100% at .8 GHz


Use case: Idle – No sleep

6/16/2014 24

We have forced the target to never

sleep!

If we don’t sleep we will not get good

battery performance

Minimizing wakeups is important but it is critical to

remember that if the system is not going to sleep in the first

place you won’t see the wakeups!


Use case: Video playback

1080p playback

6/16/2014 25

Copyright © 2014, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others. Optimization Notice 26

Use Case: Video playback


Core 1: 2.4%

Core 0: 2.7%


Kernel Timer: 1421

User Timer: 630

IRQ 46: 111

IRQ 93: 906

Frequency almost 58.9% at 2 GHz


Use case: Video playback

6/16/2014 27

Most of the time we spend in C0 during

video playback.

Video starts here, you can filter and only see wakeups during playback.


Use case: AWS Fish tank test

6/16/2014 28

The Browser is actively drawing fish

only in the light green region.

Otherwise the frequency drops and we have a regular periods

where the processor is in

C6.


Use case: AWS Fish tank test

6/16/2014 29


Core 1: 47%

Core 0: 48%


Kernel Timer: 4347

User Timer: 4120

IRQ 46: 570

IRQ 93: 557


Use Case: AWS An3DBenchXL

6/16/2014 30

Dragon Flowers Walking

Man


Use Case: AWS An3DBenchXL

Compare OpenGL ES 1.1 with ES 2.0

6/16/2014 31


Use Case: Audio playback Mp3 Audio playback scenario

Initial view Filtered and zoomed

Freq = 47ms

VTune Amplifier can help find causes for wakeups and their frequencies


mp3-nousb-screen-off mp3-screen-off

150 wakeups/sec 197 wakeups/sec

With no usb connection intel_pmu_driver can put platform in s0i3 state

Use Case: Audio playback Audio playback –Another scenario


Check if these are within expected range are not

Expert Analysis Video Recording on CLT+ - Summary


Why is display driver waking up both the CPUs ?

Use Case: Video recording Video Recording on CLT+ - Display Driver

Wakeups

Filter+Zoom


Comparisons/Regression Tests

WAKEUPS

C6 Residency

Pn Residency

Wakelocks

VTune Amplifier Views


Comparisons Native Video Playback vs 3rd Party Application

Which is “Better”?

Native Video Playback

3rd Party Application



Which is “Better”?





Much more time spent in C0 for the 3rd part application




Comparisons Adding a new Kernel Driver

Note the Wake-up Counts

Idle Workload



Note the Wake-up Counts

New Driver



VTune Amplifier Comparison Feature

Reduce the kernel logging (printk) in driver void my_timer_callback( unsigned long data )

{

int i,j,ret;

for (i=0;i<5000000;i++) {

for (j=0;j<5000000;j++) {

sum=sum+i;

printk( "my_timer_callback called %d\n", count );

}}

// setup the timer again to fire 500ms

if (count <50) {

setup_timer( &my_timer, my_timer_callback, 0 );

ret = mod_timer( &my_timer, jiffies +

msecs_to_jiffies(500) );

if (ret) printk("Error in

my_timer_callback\n");

count++;

}

}

void my_timer_callback( unsigned long data )

{

int i,j,ret;

for (i=0;i<5000000;i++) {

for (j=0;j<5000000;j++) {

sum=sum+i;

//printk( "my_timer_callback called %d\n", count );

}}

// setup the timer again to fire 500ms

if (count <50) {

setup_timer( &my_timer, my_timer_callback, 0 );

ret = mod_timer( &my_timer, jiffies +

msecs_to_jiffies(500) );

if (ret) printk("Error in my_timer_callback\n");

count++;

}

else {

printk( "callback called %d times\n",count);

}

}



VTune Amplifier Comparison Feature

Reduce the kernel logging (printk) in driver

Reduced total C0 time


References

https://lesswatts.org

6/16/2014 44


Legal Disclaimer & Optimization Notice

INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS”. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO THIS INFORMATION INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT.

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products.

Copyright © 2014, Intel Corporation. All rights reserved. Intel, Pentium, Xeon, Xeon Phi, Core, VTune, Cilk, and the Intel logo are trademarks of Intel Corporation in the U.S. and other countries.

Optimization Notice

Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Notice revision #20110804

45

Technology

Analyze and optimize Android apps power consumption