Bartendr: A Practical Approach to Energy-aware Cellular Data - PowerPoint PPT Presentation

Bartendr: A Practical Approach to Energy-aware Cellular Data Scheduling Aaron Schulman Vishnu Navda Neil Spring Ramachandran Ramjee Calvin Grunewald Venkata N. Padmanabhan University of Maryland Microsoft Research India Kamal Jain Pralhad Deshpande Microsoft Research Redmond Stony Brook University

A moving phone experiences signal strength variations -50 signal strength (RSSI) -60 0 20 40 60 80 100 120 140 160 -70 -80 -90 -100 -110 -120 -130 0 20 40 60 80 100 120 140 160 100 meter steps 2

Signal strength affects radio power and throughput 1 2500 0.9 0.8 2000 0.7 power (mW) 0.6 signal 1500 CDF -50 0.5 -60 1000 0.4 -70 -80 0.3 -90 500 0.2 -100 -110 0.1 -120 0 0 -100 -90 -80 -70 -60 -50 -40 -30 0 0.5 1 1.5 2 throughput (Mbit/s) signal strength (RSSI) 3

Signal strength affects radio power and throughput 1 2500 0.9 0.8 2000 0.7 power (mW) 0.6 signal 1500 CDF -50 0.5 -60 1000 0.4 -70 -80 0.3 -90 500 0.2 -100 -110 0.1 -120 0 0 -100 -90 -80 -70 -60 -50 -40 -30 0 0.5 1 1.5 2 throughput (Mbit/s) signal strength (RSSI) 3

Signal strength affects radio power and throughput 1 2500 0.9 0.8 2000 0.7 power (mW) 0.6 signal 1500 CDF -50 0.5 -60 1000 0.4 -70 -80 0.3 -90 500 0.2 -100 -110 0.1 -120 0 0 -100 -90 -80 -70 -60 -50 -40 -30 0 0.5 1 1.5 2 throughput (Mbit/s) signal strength (RSSI) 3

Signal strength affects radio power and throughput 1 2500 0.9 0.8 2000 0.7 power (mW) 0.6 signal 1500 CDF -50 0.5 -60 1000 0.4 -70 -80 0.3 -90 500 0.2 -100 -110 0.1 -120 0 0 -100 -90 -80 -70 -60 -50 -40 -30 0 0.5 1 1.5 2 throughput (Mbit/s) signal strength (RSSI) 3

Energy efficiency can be improved A moving phone experiences signal strength variations. Signal strength affects communication energy. Applications can hold off until signal increases and prefetch while signal is strong. 4

Energy efficiency can be improved A moving phone experiences signal strength variations. Signal strength affects communication energy. Applications can hold off until signal increases and prefetch while signal is strong. Bartendr 4

Applications can receive when signal is strong Background sync - 5 min interval sync could be more efficient if done sometime between 4 to 6 min Streaming media - Consume buffer when the signal is weak, prefetch when the signal is strong 5

Application energy measurements Drove with a mobile power monitor connected to a Palm Pre 6

Email sync energy consumption 20 15 energy (J) 10 5 0 -105 -100 -95 -90 -85 -80 -75 -70 -65 -60 -55 -50 signal strength (RSSI) 7

Email sync energy consumption 20 15 energy (J) 10 5 ✓ ✖ 0 -105 -100 -95 -90 -85 -80 -75 -70 -65 -60 -55 -50 signal strength (RSSI) 7

YouTube energy consumption 286 180 ~ ~ 160 140 120 energy (J) 100 80 60 40 20 0 -93 -73 signal strength (RSSI) 8

Applications must schedule communication Sync Streaming Schedule wakeup Fill the buffer efficiently Problem When to schedule communication to save energy? Predict signal strength Schedule syncs Schedule streaming 9

Applications must schedule communication Sync Streaming Schedule wakeup Fill the buffer efficiently Problem When to schedule communication to save energy? Predict signal strength Schedule syncs Schedule streaming Challenge Scheduling must save more energy than it consumes. 9

Obstacles to energy efficient scheduling energy consumer consumption Bartendr Signal prediction GPS is 400 mW phone already maintains locating the phone and slow to fix, signal strength, cell id, and on a path WiFi must be in neighbor cells receive mode ( 1 D not 2 or 3 D) Sync scheduler 1 J to wake up schedule syncs wakeup and sleep minutes into the future 0.5 J to sleep 3 - 10 s of radio consider the radio’s Streaming scheduler power after power state when radio energy tail communication scheduling a stream (at least 400 mW) 10

Signal strength variation on a path 1 ... 6 -50 signal strength (RSSI) -60 0 20 40 60 80 100 120 140 160 -70 -80 -90 -100 -110 -120 1 2 3 4 5 6 -130 0 20 40 60 80 100 120 140 160 100 meter steps 11

Signal strength variation on a path 1 ... 5 6 -50 signal strength (RSSI) -60 0 20 40 60 80 100 120 140 160 -70 -80 -90 -100 -110 -120 1 2 3 4 5 6 -130 0 20 40 60 80 100 120 140 160 100 meter steps 11

Predicting signal strength with previous drives 1. Find location in a previous drive Signal strength, cell id, neighbor list 2. Look ahead for future signal strength seconds in the future 12

Predicting signal strength with previous drives 1. Find location in a previous drive Signal strength, cell id, neighbor list 2. Look ahead for future signal strength seconds in the future 12

Scheduling when to sync Wake-up, sync, schedule, sleep Uses threshold for efficient sync Schedules for either first or widest signal 20 15 energy (J) 10 5 ✓ ✖ 0 -105 -100 -95 -90 -85 -80 -75 -70 -65 -60 -55 -50 signal strength (RSSI) 13

Scheduling when to sync Wake-up, sync, schedule, sleep Uses threshold for efficient sync Schedules for either first or widest signal widest first -60 -65 signal strength (RSSI) -70 -75 -80 -85 -90 -95 -100 0 50 100 150 200 250 time (s) 13

Scheduling when to receive a stream Challenge 1 . Tradeoff between strong signal and radio tail energy 2 . Signal prediction error due to speed variations 3 . Throughput prediction error due to congestion Approach 1 . Minimize predicted energy - dynamic programming algorithm 2 . Update schedule with latest signal prediction 3 . Schedule based on remaining buffer 14

past now future 15

Evaluation methodology Simulated energy consumption of naive and scheduled syncs and streaming Several 17 km drives of throughput and signal for prediction and simulation of energy consumption Started at many points in the drive 16

Syncing simulation optimal ideal first widest 1 fraction of naive energy 0.9 0.8 0.7 0 120 240 forced delay (s) 120 240 360 480 600 prediction window (s) 17

Syncing simulation optimal ideal first widest 1 fraction of naive energy 0.9 0.8 0.7 0 120 240 forced delay (s) 120 240 360 480 600 prediction window (s) 17

Syncing simulation optimal ideal first widest 1 fraction of naive energy 0.9 0.8 0.7 0 120 240 forced delay (s) 120 240 360 480 600 prediction window (s) 17

Syncing simulation optimal ideal first widest 1 fraction of naive energy 0.9 0.8 0.7 0 120 240 forced delay (s) 120 240 360 480 600 prediction window (s) 17

Streaming simulation 1 64 kbit/s fraction of naive energy 128 kbit/s 0.8 0.6 0.4 0.2 0 120 240 360 480 600 stream length (s) 18

Related work Breadcrumbs ( A. J. Nicholson et al.) Predicts WiFi network quality for a mobile device Experiences in a 3G Network (Liu et al.) and An empirical study on 3G network capacity and performance (Tan et al.) Long term throughput at a location varies TailEnder (N. Balasubramanian et al.) and Cool-Tether (A. Sharma et al.) Batching and prefetching reduce radio energy tail

Last call Signal strength affects energy consumption Applications like sync and streaming can improve energy efficiency by deferring and prefetching Previous drives can predict signal strength without breaking the energy bank Scheduling can reduce energy consumption by up to 50 % for large workloads and 10 % for small 20