4.Tizen Power Management (Tizen 2.3)Tizen_2... · Power Management in Tizen and Linux Kernel (Tizen 2.3) ... OS Power Management Hardware: CPU, BIOS etc. ... Low processor frequency

Real-Time Computing and Communications Lab., Hanyang Universityhttp://rtcc.hanyang.ac.kr

Real-Time Computing and Communications Lab., Hanyang Universityhttp://rtcc.hanyang.ac.kr

Power Management in Tizenand Linux Kernel (Tizen 2.3)

Minsoo Ryu

Real-Time Computing and Communications Lab.Hanyang University

[email protected]

2Real-Time Computing and Communications Lab., Hanyang University

http://rtcc.hanyang.ac.kr 2Real-Time Computing and Communications Lab., Hanyang Universityhttp://rtcc.hanyang.ac.kr

Outline

Introduction ACPI (Advanced Configuration Power Interface) DVFS (Dynamic Voltage Frequency Scaling) Power Management in Tizen Lab Assignment



Why Power Management?

Battery-operated devices Smartphones, digital cameras, and laptops use batteries Power savings and battery run times are paramount

Increasing energy use of servers Servers and data centers in U.S. reached 1.5% of the

national electricity consumption in 2006 The electricity consumption of servers doubled between

2000 and 2006



Power Consumption in Smartphones



Power Consumption of Memory

In a server system Memory consumes 19% of system power on average Some work notes up to 40% of total system power



Idle Power Consumption

Only 30% of servers in data centers are fully utilized while keeping the other 70% in idle state Idle servers consume between 60% and 66% of the peak

load power consumption



Two Dimensions of Power Management

Power management when the system is idle Select the most efficient idle state

Power management when the system is active Dynamically change operating frequency and/or voltage



ACPI(Advanced Configuration Power Interface)



APM and ACPI

APM (Advanced Power Management) Activated when system becomes idle

• e.g.) screen saver => sleep => suspend Controlled by firmware (BIOS)

• need reboot for reconfiguration OS has no knowledge

ACPI (Advanced Config. and Power Interfaces) Controlled by OS First released in 1996 by Compaq/HP, Intel, Microsoft, …



ACPI

Standard interface specification Brings power management under the control of the

operating system The specification is central to Operating System-directed

configuration and Power Management (OSPM)

OS Power Management

Hardware: CPU, BIOS etc.

Software driversACPI

Applications



ACPI Functions

System power management The entire computer

Processor power management When OS is idle but not sleeping, it puts processors in low-

power states

Device power management ACPI tables describe motherboard devices, their power

states, the power planes the devices are connected to



Firmware-Level ACPI Architecture

Three components ACPI tables

• Contain definition blocks that describe all the hardware that can be managed through ACPI

• Include both data and machine-independent byte-code• OS must have an interpreter for the AML bytecode

ACPI BIOS• Performs basic management operations on the hardware• Include code to help boot the system and to put the system to

sleep or wake it up ACPI registers

• A set of hardware management registers defined by the ACPI specification



Firmware-Level ACPI Architecture



ACPI States



Global States

G0: Working (S0) Processor power states (C-state): C0, C1, C2, C3

G1: Sleeping (e.g., suspend, hibernate) Sleep State (S-state): S0, S1, S2, S3, S4

G2: Soft off (S5) Almost the same as G3 Mechanical Off, except that the

power supply unit (PSU) still supplies power at a minimum Other components may remain powered so the computer

can "wake" on input from the keyboard, clock, modem, LAN, or USB device

G3: Mechanical off



Processor States (C-state)

Global state is G0 (working) Four processor states

C0: Operating• Performance state (P-State) • P0: highest performance, highest power • P1 ~ Pn: lower performance, lower power

C1: Halt• The processor is not executing instructions, but can return to an

executing state essentially instantaneously C2: Stop-Clock (optional)

• The processor maintains all software-visible state, but may take longer to wake up

C3: Sleep (optional)• The processor does not need to keep its cache coherent, but

maintains other state



Performance States (P-State)

Intel Pentium M at 1.6 GHz



Device States (D-State)

The device states D0–D3 are device-dependent D0: Fully On

• The operating state D1 and D2

• Intermediate power-states whose definition varies by device D3: Off

• The device is powered off and unresponsive to its bus• D3 Hot: Aux power is provided• D3 Cold: No power provided



Sleeping States (S-State)

Four sleeping states S1: Power on Suspend (POS)

• All the processor caches are flushed• The power to the CPU(s) and RAM is maintained• Wakeup takes about 1 ~ 2 seconds on desktops

S2: CPU powered off• Dirty cache is flushed to RAM (Often not used)

S3: Suspend to RAM (STR), or Standby, Sleep• RAM remains powered• Wakeup takes about 3 ~ 5 seconds on desktops

S4: Suspend to Disk (STD) or hibernation• All content of the main memory is saved to non-volatile memory

such as a hard drive, and is powered down



DVFS(Dynamic Voltage Frequency Scaling)



Power Consumption in CMOS

Power consumption per every cycle in a digital circuit C (output capacity) V (supply voltage) f (clock frequency)

Energy consumption 2VE



Power Consumption in CMOS

Relationship between V and f

Lowering the supply voltage increases the circuit delay The clock frequency must be reduced accordingly

Observations Execution times scale linearly to the supply voltage The energy consumption scales quadratically to the supply

voltage

fV



Dynamic Voltage Frequency Scaling

Voltage scheduling graph by Pering, et al. {S, C, D} = {start time, execution time, deadline} {E} = energy

Optimal speed settings Sum-of-squares optimization Given the constraint

• is minimized when

•

kfi

2if

nfff ...21



Four Considerations for DVFS

Workload amount Adjust the processor frequency depending on the load

Workload characteristics Compute-intensive vs. memory-intensive

Deadline constraints Lowest possible frequency for meeting deadlines

Load balancing Migrate or scale?



Workload Amount and DVFS

Static approaches Performance policy

• CPU runs at the maximum frequency regardless of load Power save policy

• CPU runs at the minimum frequency regardless of load

Dynamic approaches On demand policy

• Increase the clock speed to the maximum frequency when the system load goes above the predefined threshold

• Decrease the clock speed gradually when the system load becomes below the predefined threshold

Conservative policy• Gracefully increase the CPU speed rather than jumping to the

maximum speed



Workload Characteristics and DVFS

Two types of workload Compute-intensive

• The program execution is exclusively bound to the processor Memory-intensive

• The program makes heavy access to memory• The processor would spend a significant fraction of the time waiting

for memory

A simple solution High processor frequency and low memory frequency for

compute-intensive load Low processor frequency and high memory frequency for

memory-intensive load



CPU- vs. Memory-Intensive

Execution time variation CPU frequency ranging from 733 MHz to 333 MHz



GPU- and Memory-Intensive

Compute-intensive applications Dense matrix multiplication Run on NVIDIA GeForce GTX 280 GPU



GPU- and Memory-Intensive

Memory-intensive applications Dense matrix transpose Run on NVIDIA GeForce GTX 280 GPU



Deadline Constraints and DVFS

The CPU runs at “just-enough” speed to meet deadline constraints

timedeadlinearrival

maxfreq.

timedeadlinearrival

maxfreq.

minfreq.

minfreq.



Load Balancing and DVFS

DVFS can be independently applied to each processor on multicore hardware But this may not lead to optimal power saving from a global

point of view

A simple scenario We need to decide whether to transfer a thread from

processor A to an idle processor B, or increase the frequency of A

Compute Pmigrate_from_A_to_B and Pincrease_A_freq

Transfer if Pmigrate_from_A_to_B < Pincrease_A_freq

Otherwise, increase the frequency of A



DVFS Overhead (Intel Core2 Duo E6850)

uc (underclocking) ir (inductor loss)



DVFS Overhead (Exynos 4210)



POWER MANAGEMENT IN TIZEN V2.3



Overview of Tizen’s Power Management(Tizen 2.3)



Power Management in Tizen 2.3

User space applications can request a specific power state to Power Manager



Power Manager

Power manager controls LCD and system operation state which are relevant with power according to configurations

There are 5 power states in power manager START NORMAL LCDDIM LCDOFF SLEEP



Power Manager

Transition procedure is done as follow steps1. Call check2. Transition3. Call enter action function

In the call check phase, check transition to next state is possible

If it is possible, change the current state to the next state according to transition table and do the entering action



Power Manager

The states are changed by input or timeout event as shown in previous diagram

Transition table indicates next states

State Timeout InputSTART START START

NORMAL LCDDIM NORMALLCDDIM LCDOFF NORMALLCDOFF SLEEP NORMALSLEEP LCDOFF NORMAL



Power Manager

In each state’s action function, it does corresponding operation NORMAL: backlight on and restore the previous brightness LCDDIM: dim the brightness LCDOFF: turn off the backlight SLEEP: set system mode to suspend

Power manager changes not only backlight state but system operation state properly(i.e. suspend, pre_suspend, post_resume)



Sleep State: Suspend-to-RAM

Suspend-to-RAM1. Sync File system / Freeze user processes2. Suspend devices3. Stop IRQ4. Suspend system core5. Configure sleep / Save PC

Waking-up1. Wakeup Interrupt2. CPU boot-up / load bootloader3. Bootloader Restore PC4. Conduct operations in reverse order of Suspend-to-RAM



Sleep State: Suspend-to-RAM



Power Management in the Linux Kernel

Tizen is based on Linux kernel The linux power management techniques are applied Idle state

• ACPI – sleeping states Active state

• CPUfreq, Devfreq• DVFS governor

Tizen kernel can also be modified & ported to device separately



Power Management in the Linux Kernel



LAB 2: IMPLEMENTING AN ADDITIONAL POWER STATE IN

TIZEN



Lab Assignment

Implement an additional power state, SEMIDIM, in power manager SEMIDIM state is an intermediate state between NORMAL and

LCDDIM SEMIDIM state sets the backlight brighter than LCDDIM

• Half of NORMAL brightness SEMIDIM state lasts for the same time duration as LCDDIM



Lab Assignment

START

NORMAL

LCD_DIM

LCD_OFF

SLEEP

SEMIDIM

Notification from Launcher UITimeout eventInput device eventKernel / HW



Lab Assignment

START

NORMAL

LCD_DIM

LCD_OFF

SLEEP

SEMIDIM

Notification from Launcher UITimeout eventInput device eventKernel / HW

The new part to be implemented



Lab Assignment

State transition table with SEMIDIM

Other states are not changed except NORMAL state’s timeout transition

State Timeout InputNORMAL SEMIDIM NORMALSEMIDIM LCDDIM NORMALLCDDIM LCDOFF NORMALLCDOFF SLEEP NORMAL



Power Manager

In Tizen 2.3, power manager is integrated in display control Power manager code is in the display component of deviced framework/system/deviced/display



Power Manager

Get source code

Change the branch to Tizen version 2.3



Power Manager

The main code is core.c

It includes the definition of power states and power management operations



Power Manager

In display_init(), the power manager is initiated with backlight and power operations

Register setting callback

Set the timeout duration

Initiate polling function

dbus for api



Power Manager

Each state has 3 functions for transition int (*trans) (int evt): transition function int (*action) (int timeout): enter action int (*check) (int next): transition check function

In the trans function, the check and action functions are called as subroutines

All the states have the same default functions default_trans(), default_action(), default_check()



Power Manager

State transitiontriggered

default_trans()

default_check()

default_action()

Check the next stateis available

Do the correspondingactions of the changed state



Power Manager

In default_action(), the backlight and system power states are changed as the current state

Backlight operations are called to change backlight brightness S_NORMAL calls update() to light up the display S_LCDDIM calls dim() to make the display dim

The proper operation for SEMIDIM state should be implemented



Power Manager

Timeout value for S_NORMAL and S_LCDDIM is determined by entire LCD on duration S_NORMAL

S_LCDDIM

SEMIDIM state last same time with LCDDIM, and entire LCD on time should be properly distributed



Power Manager

By the key-filter, the inputs are ignored if the LCD is off state

Therefore, S_SEMIDIM also should be included in LCD on state

Documents

4.Tizen Power Management (Tizen 2.3)Tizen_2... · Power Management in Tizen and Linux Kernel (Tizen 2.3) ... OS Power Management Hardware: CPU, BIOS etc. ... Low processor frequency