Energy-Delay Tradeoffs in Smartphone Applications

1. MobiSys’10 Energy-Delay Tradeoffs in Smartphone Applications Moo-Ryong Ra JeongyeupPaek, Abhishek B. Sharma RameshGovindan, Martin H. Krieger, Michael J. Neely University of Southern California

2. The Urban Tomography System • Video Collection 10 years ago • Our way of Video Collection

3. Documenting • Post-Katrina • Reconstruction Users • Transportation Security • Post-Disaster Urban Planning

4. Delay-Tolerance • Many of our users are delay-tolerant • But tolerance varies considerably Planning Research Transportation Security Delay-Tolerance Dealing with Child development Issues

5. Our Focus Transferring Large Volumes of Data Leveraging Delay Tolerance Reduce the energy cost

6. Trade-offs • Delay transmission • Adapt to wireless • channel quality HIGH HIGH Time-Varying LOW LOW Time-Varying

7. A Motivating Example EDGE 3G WiFi 200 KB/s video 2 arrives 50 KB/s video 1 arrives 10 KB/s 40 KB/s TIME

8. Strawman Approaches EDGE 3G WiFi • Energy-Optimal • WiFi-Only • Min-Delay Optimal can save significant energy Challenge: How to design the optimal trade-off algorithm? video 2 video 1 TIME Delay Energy 246 320 305 sec 242 J 95 50 MD ME EO MD ME EO

9. Our Approach Use Lyapunov Optimization Derive Control Algorithm

10. Lyapunov Optimization • 1. Queue Stability • 2. Penalty Minimization Sender Receiver Queue length will not go to the infinity Lyapunov Analysis Control Algorithm Penalty Function

11. How we use the framework • 1. Queue Stability • 2. Penalty Minimization • 1. Delay Bound • 2. Energy Minimization Sender Receiver Lyapunov Analysis Control Algorithm Energy Cost Model

12. Derived Control Decision SALSA(Stable and Adaptive Link Selection Algorithm) Queue Backlog Estimated Rate V Energy Cost MAX − × × Over All Links AP Higher rate Q Large Queue Backlog E Low Energy Cost

13. Importance of V SALSA(Stable and Adaptive Link Selection Algorithm) Queue Backlog Estimated Rate V Energy Cost MAX − × × Over All Links Setting V Large Defer the Transmission

14. Tuning Delay-Tolerance • Transportation Security • Urban Documentation AP Rate TIME Transmit? No No Yes No No Yes V Queueing Delay

15. SALSA Summary • SALSA Whether, When, Which Tunable Delay-Tolerance Delayed Transmission

16. Evaluation Methodology • USC • LAX • Mall Trace-Driven Simulation 2772 66 Link availability traces 42 Video arrival traces Implementation on Nokia N95 Validation with Real Implementation

17. Metric • Envelope in E-D Plane • Dispersion BAD Measured D(Hour) D(Hour\byte) dispersion E(J/byte) E(J/byte) Good! Optimal

18. Comparisons Min-delay WiFi-only NOT take link quality into account NOT consider queue backlog Unbounded delay High Energy Static-delay Know-WiFi

19. Main Results Ignore queue backlog Since SALSA takes all factors into account, it performs closest to the optimal Ignore link quality Min-Delay WiFi-only Static-Delay Know-WiFi SALSA

20. Tuning Delay-Tolerance Like WiFi-only Like Min-Delay Decay V slowest Decay V fastest SALSA can be tuned to different delay-tolerant requirements. More Delay-Tolerant Less Delay-Tolerant

21. Experimental Validation • Glendale • USC Simulation Results Experiment Results Our experimental results fall within the boundary and validate the simulation

22. The Bottom Line • Additional Delay • Energy Savings Save 2% ~ 80% of battery capacity • Loss? • Gain? + 2 min ~ 2 hour • Battery

23. Summary of Contributions • Adaptive algorithm for energy/delay tradeoff • Extensive evaluation with real world scenarios • Validation with real implementation • Provable performance bound (in the paper) • Different from prior work • [Balasubramanian10], [Zaharia07], [Seth06] consider schemes similar to Static-Delay and Know-Wifi. • [Rahmati07], [Armstrong06], [Agarwal07] consider link selection, but do not defer transmissions

24. Thank you. Questions?