1 / 11

Energy Efficient Computing: 3 observations and 3 lessons from embedded systems

Energy Efficient Computing: 3 observations and 3 lessons from embedded systems. Engine. Work. Computer. Information. Heat. EM energy. Heat. Heat. Rajesh Gupta, UC San Diego http://mesl.ucsd.edu. Microsoft, July 2009. <1MIPS/mW. >1MIPS/mW. GIPS. MIPS.

sharis
Download Presentation

Energy Efficient Computing: 3 observations and 3 lessons from embedded systems

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. Energy Efficient Computing: 3 observations and 3 lessons from embedded systems Engine Work Computer Information Heat EM energy Heat Heat Rajesh Gupta, UC San Diego http://mesl.ucsd.edu Microsoft, July 2009

  2. <1MIPS/mW >1MIPS/mW GIPS MIPS Large Differences between components 10-100W mW-100mW O1: Heterogeneity is a fact of life.

  3. Cellular voice radio (GSM) highly optimized for low idle power • Cingular 2125: GSM radio consumes 38 times less power than Wi-Fi ! Power breakdown for a fullyconnected mobile device in idle mode, with LCD screen and backlight turned off. 6-10X variation in power use is common. O2: Increasing bandwidth of power consumption.

  4. Desktop PC Active State : >140W Idle State : 100W Sleep state : 1.2W Hibernate : 1W Large Differences between ON & OFF O3: Abstraction stacks cost power.

  5. One Lesson in Three Parts • Exploit Heterogeneity: large differences • Between components, power states • Deliver power when and where needed • JIT power delivery • Match application needs to power availability • Differentiated quality of computation Beyond all low power tricks, duty-cycle keeps on giving

  6. Bluetooth Wi - Fi WiFi WiFi WiFi BT WiFi Active Active Active BT Active Active WiFi Sniff PSM 5.8 mW 81 mW 264 mW 990 mW Easier Said Than Done wakeupfilters Appln. stubs • Reliability? Availability? Usability? Embedded OS, including networking stack Paging Radios Embedded CPU, RAM, flash Host PC Apps Somniloquy daemon Operating system, including networking stack Host processor, RAM, peripherals, etc. Secondary processor Sleep Talking Processors Network interface hardware Exploit heterogeneity to lower power by duty cycling.

  7. Somniloquy USB Interface (Wake up Host + Status + Debug) USB Interface (power + USBNet) SD Storage Processor 100Mbps Ethernet Interface In collaboration with Microsoft Research

  8. 92% less energy than using host PC. 1 600 1200 1800 2400 Increase battery life from <6 hrs to >60 hrs • From Embedded Systems to HPC: We KNOW duty cycling is useful.

  9. Takeaways • Slowdown or low power design ultimately reaches a limit • Duty-cycling keeps on giving • But causes non-trivial problems in availability, usability, reliability • Challenge for the community • Algorithms: what are the right combination of slowdown and shutdown strategies? • Architectures: what is the right organization of components for maximal duty cycling? “Future lies in system architectures built for aggressive duty-cycling”

  10. “Koala Class” Computing “Size Your Brain Power, Storage, and Sleep Cycles to Your Problem,” Tom DeFanti

More Related