1 / 30

Software Strategies for Portable Computer Energy Management

Software Strategies for Portable Computer Energy Management. IEEE Personal Communications 1998 Presented by Hsu Hao Chen. Outline. Introduction General Issues Strategies for Components Secondary storage Processor Wireless communication devices Display and backlight Conclusions.

Download Presentation

Software Strategies for Portable Computer Energy Management

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Software Strategies for Portable Computer Energy Management IEEE Personal Communications 1998 Presented by Hsu Hao Chen

  2. Outline • Introduction • General Issues • Strategies for Components • Secondary storage • Processor • Wireless communication devices • Display and backlight • Conclusions

  3. Introduction(1/2) Limiting the energy consumption of computers, especially portables, is becoming increasingly important. • Goal • reduce energy consumption is to simply use components thatconsume less power. • use components thatcan enter low-power modes by temporarily reducing theirspeed or functionality • discuss software techniques for taking advantages of low-power hardware

  4. Introduction(2/2) • Power-saving three categories of software strategies • transition issue • load-change issue • adaptation issue

  5. Introduction • General Issues • Strategies for Components • Secondary storage • Processor • Wireless communication devices • Display and backlight • Conclusions

  6. Strategy Types • Transition strategies (prediction strategies) • determining when to switch to low-power modes • knowledge about its mode characteristics • information about its future functionality requirements • Load-change strategies • how to modify the load on a component in order to increase its use of low-power modes • Adaptation strategies • how to create software that allows components to be used in power-saving ways

  7. Application-aware adaptation User Level OS Level Component Level Application Level Levels of Energy Management Lack certain useful informationabout the state ofthe machine Abstraction Lackinformation about the overall workload

  8. Strategy Evaluation and Power Budget • Strategy Evaluation • Power Budget mechanical motion and light generation seem inherently more power intensive than digital circuits.

  9. Introduction • General Issues • Strategies for Components • Secondary storage • Processor • Wireless communication devices • Display and backlight • Conclusions

  10. standby r/w r/w r/w active sleep idle off control control control control cache cache cache cache cache spin spin spin spin spin r/w r/w control Hardware Features • Most hard disks have five power modes:

  11. Transition Strategies • Disks can be of three kinds: • deciding when to go to sleep mode • deciding when to go to standby mode (no studies) • deciding when to go to off mode • Common strategy • use inactivity thresholds as low as 1-10 seconds (Researchers discovered ) • enter sleep mode after a fixed inactivity threshold (3-5min) • Problems • make the user wait more often for the disk to spin-up • cause disk failure from frequent spin up/down cycles • It is very difficult to predict accurate disk access pattern due to wide-varying of human activities

  12. Changing Inactivity Threshold • First argument • assumption: disk access interarrival times are drawn from some unknown stationary distribution • keep track of all interarrival times in order to deduce the best threshold • Second argument • assumption: disk access interarrival time distribution is nonstationary (i.e. changing with time) • adapt its threshold based on recent samples • Third argument • assumption: make no assumption about disk access pattern • choose thresholds randomly

  13. Load-change Strategies Common strategy • changing the configuration or usage of the disk cache • increase the cache size • changing the dirty block timeout period (0 to 30s) • add file names and attributes caching Prefetching • disk cache is filled with data thatwill likely be needed in the future before it is spun down Reduce paging activity • reduce working set sizes • improve memory access locality

  14. Adaptation Strategies(1/4) • Flash memory • advantage: • nonvolatile and low power consumption • reading as fast as DRAM • no seek latency • disadvantage: • poorer write performance (10-100 times slower than hard disk) • data cannot be overwritten without erasing the entire segment containing it

  15. Adaptation Strategies(2/4) • Flash memory as disk cache • effective as a second-level cache • most writes would beflushes from first-level and asynchronous • problem: • cannot be overwritten without erasing the segment • sol: • ensure there is always a segment with free space for writing

  16. Adaptation Strategies(3/4) • Flash memory as disk • No seek latency • disk cache is no longer important • except to be used as a write buffer (asynchronous write) • problem: • cannot be overwritten without erasing the segment • sol: Log-structured file system • new data does not overwrite old data but appended to a log

  17. Adaptation Strategies(4/4) • Wireless network as disk • no power consumption of the storage device • problem: • increased power consumption of the wireless communication • higher latency for file system accesses (wireless bandwidth) • Several models • transmit data access requests to, and receive data from, a server • storage device as a large cache for the server file system • perform all processing on a server The limiting factor in all of these cases is networkbandwidth

  18. Introduction • General Issues • Strategies for Components • Secondary storage • Processor • Wireless communication devices • Display and backlight • Conclusions

  19. Hardware Features Slow down the clock • without changing the voltage is not useful • energy consumption is invariant with clock speed Selectively shut off functional units • such as the floating-point unit Shut down processor operation altogether • return to full power when the next interrupt occurs

  20. Transition Strategies • When the processor can turn off • from the current status of all processes • when all processes are blocked • When it can change the processor speed • slowing the clock is accompanied by reducing the voltage problem: • make an appropriate trade-off between energy reduction and performance • Adjust the processor speed gradually • interval-based approach attempting to complete all processor work by the end of each interval

  21. Load-change Strategies • When the processor can turn off • reduce the time tasks take (efficient OS) • use lower-power instructions (energy aware compilers) • reduce the number of unnecessary tasks • whenan application is idle, it will “busy-wait” for an event insteadof blocking (period satisfies conditions) • When functional units can turn off • compiler that clusters together several uses of a pipelined functional unit • compiler that generates instructions using functional units that do not get power-managed

  22. Introduction • General Issues • Strategies for Components • Secondary storage • Processor • Wireless communication devices • Display and backlight • Conclusions

  23. Hardware Features • Five operating modes: • transmit • receive • idle • no tx/rx • transceiver is still powered and ready to rx or tx • sleep • similar to idle but transceiver powered down • off • Change transmission power • Power consumption depends on the distance to the receiver

  24. Transition Strategies • When to enter sleep mode • strategies similar to those for hard disks • inactivity threshold methods • wireless device does not have the large mechanical components • wireless device should periodically exit sleep mode to send signal to the server (alive) • Effects by changing transmission power • increased battery lifetime • lower bit error rate for neighbors • higher bit error rate for one’s own transmissions • When to change power • local strategy • global strategy

  25. Load-change Strategies • When sleep mode is used • increase the amount of time it can spend sleeping • reduce network usage altogether • Ex: compress TCP/IP headers; stop transmission when channel is bad • Use MAC protocol • sleep when it is certain that no data will arrive for it • each mobile unit only be awake during the broadcast of the traffic schedule • When it can change transmission power • mitigate the side effects of increased bit error rate

  26. Introduction • General Issues • Strategies for Components • Secondary storage • Processor • Wireless communication devices • Display and backlight • Conclusions

  27. Hardware Features • Energy-saving features • Backlight • brightness level • turn it off • Display • switch from color to monochrome • reduce update frequency • turn it off

  28. Software Strategies • When to turn off • after a certain period of time has passed with no user input • make it progressively dimmer • When to switch mode or reduce frequency • Current display mode are not visually important to the user

  29. Introduction • General Issues • Strategies for Components • Secondary storage • Processor • Wireless communication devices • Display and backlight • Conclusions

  30. Conclusions • Power-saving three categories of software strategies • transition • load-change • Adaptation trade-off between energy savings and performance • hardware features are rarely complete • not only reducing the power consumption, but also introducing lower-power, lower-functionality modes

More Related