Rethinking OS Design

1. Rethinking OS Design Productivity applications Process control Personal (PDAs), Embedded Workload You are here Services & API Internal Structure Metrics Policies / Mechanisms Energy efficiency Hardware Resources Processor, Memory, Disks (?), Wireless & IR, Keyboard(?), Display(?), Mic & Speaker, Motors & Sensors, GPS, Camera, Batteries

2. Energy Efficiency Metrics • Power consumption in watts (mW). • Battery lifetime in hours (seconds, months). • Energy consumption in Joules (mJ). • Energy * Delay • Watt per megabyte

3. Physics Basics • Voltage is amount of energy transferred from a power source (e.g., battery) by the charge flowing through the circuit. • Power is the rate of energy transfer • Current is the rate of flow of charge in circuit

4. Relationships Energy (Joules) = Power (watts) * Time (sec) E = P * t Power (watts) = Voltage (volts) * Current (amps) P = V * I Current (amps) = Voltage (volts) / Resistance (ohms) I = V / R

5. Battery Terminology • Primary (non-reusable) and Secondary (rechargable) • Voltages: Voc (initial no-load)V (operating voltage under load)Vcut (cut-off when cell is considered discharged - 80% of Voc) • Capacity expressed in amp-hourstheoretical - based on amount of material in cellnominal - based on amp-hours obtained when discharged at constant current until Vcut

6. Battery Terminology • Discharge time - elapsed time until a fully charged cell reaches Vcut • C rate - discharge current expressed in amps relative to nominal capacity • example: for a lead acid battery with nominal capacity of 5Ah, a discharge rate of C/20 means 250mA of current. • Specific energy - Watt-hours per kilogram delivered at constant discharge • Energy density of cell - Watt-hours per liter

7. Battery Technology

8. Discharge Behavior Discharge behavior of lithium-ion cell withVoc = 3V and Vcut = 1V

9. Battery Stuff • Diffusion: At non-zero current, active material at electrode-electrolyte interface are consumed and replaced by new stuff moving in • Polarization as current increases; At high enough current, diffusion is unable to compensate for depletion at electrode and cell voltage drops • Recovery (due to diffusion) when current decreased

10. Ragone plot for different chemistries

11. Pulsed Discharge • Exploiting recovery ability to get more out of a battery • Delivered specific energy can be increased by pulsed instead of constant discharge for a fixed power level. • [Chiasserini and Rao 99] - model & analysis • Is bursty better for battery lifetimes? • Can durations of idle and busy states be optimized?

12. Pulsed Discharge Bipolarlead acid cell Pulse = 3msRest = 22ms

13. Smart Batteries • Part of Intel Power Initiative • Embedded battery controller that can be controlled by OS. • Interface • Battery reports designed capacity, latest full charged capacity, remaining capacity. • Warning levels can be set. User notifications

14. Rethinking OS Design Productivity applications Process control Personal (PDAs), Embedded Workload Services & API Internal Structure Metrics Policies / Mechanisms Energy efficiency You are here Hardware Resources Processor, Memory, Disks (?), Wireless & IR, Keyboard(?), Display(?), Mic & Speaker, Motors & Sensors, GPS, Camera, Batteries

15. System Organization interrupts Processor Cache Memory Bus I/O Bridge I/O Bus Main Memory Disk Controller Graphics Controller Network Interface Graphics Disk Disk Network

16. Power Budgets ave 18% interrupts Processor Cache Memory Bus I/O Bridge I/O Bus Main Memory 4-17% ave. 9% +backlight 23% Disk Controller Graphics Controller Network Interface Graphics Disk Disk Network [Lorch95] appox 20% 4-12% ave. 8%

17. Typical Notebook Power Budgets (Color to 21W) 8 [Harris 95] B/W 6 DC-DC HDD Watts 4 video mem 2 CPU 1993 notebook full power

18. What are the Costs?Measured Power Consumption (PalmPilot Pro - 1997 model) Hotsync Backlight Memory intensive CPU Event Loop (nilevents) CPU Idle Sleeping in cradle

19. [Tiwari94] 486DX2 Instr current (mA) NOP 276 Load 428 Store 522 Register add 314 cache miss 216 Memory op current (mA) no access 5-77 page hit 123 page miss 248 CPU/Memory

20. Intel Power Initiative Targets

21. Power Budget Targets 33% interrupts Processor Cache Memory Bus I/O Bridge 10% I/O Bus Main Memory 13% Disk Controller Graphics Controller Network Interface Graphics Disk Disk Network 8% 2- 3% [Intel targets] 4%

22. Itsy Measurement Methodology Isense = Vsense /.02 Sampling rate: 5000 per second

23. Itsy Results

24. PowerScope [Flinn] • Statistical sampling approach • Program counter/process (PC/PID) + correlated current readings. • Off-line analysis to generate profile • Causality • Goal is to assign energy costs to specific application events / program structure • Mapped down to procedure level • System-wide. Includes all processes, including kernel

25. Experimental SetupData Gathering Multimeter’sclock drivessampling at period of 1.6ms Takescurrentsample -> InterruptcausesPC/PID sample to be buffered ->Trigger next sample <-TriggerProfilingcomputer User-level daemonwrites to disk when buffer 7/8 full

26. System Monitor Kernel Mods • NetBSD • recording of PC and PID • fork(), exec(), exit() instrumented to record pathname associated with process • new system calls to control profiling • pscope_init(), pscope_start(), pscope_stop(), pscope_read() (user-level daemon, to disk)

27. Energy Analyzer • Voltage essentially constant, only current recorded. • Each sample is binned into process bucket and procedure within process bucket. • Energy calculated by summing each bucketE = Vmeas S ItDt n t=0

29. Case Study Video applicationoriginal 12.1MB • Step 1lossy compressionB: 7MB, C: 2.8MB • Step 2: display size reduced from 320x240 to 160x120Asmall: 4.9MB, Csmall: 1MB • Step 3: WaveLAN put into standby mode when not used • Step 4: Disk powered off

30. Base case Every optimization

31. Energy = S Powerix Timei To reduce energy used for task: Reduce power cost of power state ithrough better technology. How to Reduce Energy Consumption? i e powerstates

32. Opportunities for Lower Power through Technology Circuits • Gated Clocks - disable functional units that are not in use for particular instruction • Compile for • Voltage Scaling For given circuit: • E is related to V2 and time, f(clockrate) • Linear relationship between V and clockrate • Ability of software to dynamically change?

33. Displays • Active Matrix LCDs • 90% of backlight gets transmitted through the layers of display • Possible future technologies • Reflective displays use ambient light 1/50th energy of active matrix • Field-emission displays uses an array of cathodes for each pixel instead of one gun as in CRT displays. Selective activation possible.

34. Energy = S Powerix Timei To reduce energy used for task: Reduce power cost of power state ithrough better technology. Reduce time spent in the higher cost power states. How to Reduce Energy Consumption? i e powerstates

35. Power Modes of HW Devices High power cost transition Busy ? Idle Low power cost transition ?

36. Energy = S Powerix Timei To reduce energy used for task: Reduce power cost of power state ithrough better technology. Reduce time spent in the higher cost power states. Amortize transition states, if significant. How to Reduce Energy Consumption? i e powerstates

37. StrongARM Processor Power Modes 160MHz microprocessor, 2 16kB caches on chip • Normal active mode: 450mW • Idle mode: 20mW, return to normal, no delay. • Internal clocking stopped • Sleep mode: 150mW, return to normal 140ms • Internal power to chip off, I/O circuitry remains powered, no state saved

38. Rambus RDRAM Power Modes Read/write transaction Active 1.0x mW 100x ns 1.0x ns PwrDown .01x mW Nap 0.1x mW 0.1x ns Standby 0.6x mW

39. spinup

40. Wireless LAN Power Modes

41. Listen every 1.28 sec. Bluetooth • Freq Hop Radio • nominal range 10 meters • augmentable to 100 meters with power amplifier • 721 kbits/sec • Adaptive range-RSSI (received signal strength indicator)