1. Power Scheduling at the Network Layer for wireless sensor networks Barbara Hohlt Eric Brewer UC Berkeley NEST Retreat June 2004

2. Wireless sensor networks Lifetime constrained by limited energy stores Communication is the dominant energy cost Turning the radio off during idle times reduces power consumption Flexible Power Scheduling  Adaptively schedules nodes to save radio power  Decentralized  Multihop sense-to-gateway applications  Typically 5X power savings

3. Time in seconds Current in mA 1.4 Overall: 5X Savings

4. Six Design Principles Avoid idle listening Use a schedule Two-layer architecture Schedule traffic flows (not packets) Schedules must be adaptive Nodes that want change do the most listening

5. Power Schedule MAC FPS Two-Layer Architecture Coarse-grain slotted scheduling  At the network layer  Plan radio on-off times Fine-grain CSMA MAC underneath Reduces contention and increases end-to-end fairness  Distribute traffic  Decouple correlated events from traffic  Reserve bandwidth from source to sink Does not require perfect schedules or precise time synchronization

6. FPS Experiments 10 MICA motes plus base station 6 motes send 100 messages across 3 hops One message per cycle (3200ms) Begin with injected start message Repeat 11 times 123456 Two Topologies  Single Area one 8’ x 3’4” area  Multiple Area five areas, motes are 9’-22’ apart Scheduled (FPS) vs Unscheduled (Naïve)

7. Contention is Reduced

8. End-to-end Fairness FPS Naive AVGSTDDEVMax/Min FPS96.41.131.03 Naïve24.76.192.48

9. Evaluation with TinyDB Three implementations  TinyDB duty cycling  TinyDB low power listening  TInyDB FPS Berkeley Botanical Gardens

10. 3 Step Methodology Estimate radio-on time for each scheme For FPS, validate the estimate at one mote Use current measurements to estimate power consumption

11. TinyDB Redwood Deployment 17 18 BTS 12 3 0 2 trees 35 nodes 1/3 two hops 2/3 one hop No radio power management = 3600 sec/hour

12. TinyDB Duty Cycling 4 seconds 2.5 minutes All nodes wake up together for 4 seconds every 2.5 minutes. During the waking period nodes exchange messages and take sensor readings. Outside the waking period the processor, radio, and sensors are powered down. 24 samples/hour * 4 sec/sample = 96 sec/hour

13. Low-Power Listening Radio-on time = listening + transmitting + receiving.003 sec/poll * 10 polls/sec * 3600 sec/hour= 108 sec/hour to listen ( 24 samples/hour ) * ( 2/3 * 1 hop + 2/3 * 1 hop ) = 32 hops/hour 32 hops/hour * 0.1 sec/hop = 3.2 sec/hour to transmit 108 (L) + 3.2 (T) + 1.6 (R) = 112.8 sec/hour

14. Flexible Power Scheduling 18 slots * 128 ms = 2.3 sec/cycle per 3 nodes = 0.767 sec/cycle (per node) 24 samples/hour * 0.767 sec/cycle = 18.4 sec/hour 0 2 3 1 Traffic Comm Node 1: 2 T, 3 A Node 2: 3 T, 2 R, 3 A Node 3: 2 T, 3 A 5 (node 1) + 8 (node 2) + 5 (node 3) = 18 slots

15. FPS Validation

16. Power ratios: 160x 4.4x 5.1x 1

17. Summary Flexible Power Scheduling  Two-level architecture  Schedules flows (not packets)  Adaptive and decentralized schedules Reduced contention and increased end-to- end fairness and throughput Improved power savings of 4.4X over duty cycling and 160X over no power management

18. Thank You Barbara Hohlt hohltb@cs.berkeley.edu

19. Radio-on Times Radio on: 8 mA Radio off and node on: 0.4 mA Radio off and node asleep: 0.01 mA

20. Power Savings Radio Off and Node Asleep Radio Off but Node On (Worst Case)

21. END