CS 410/510 Sensor Networks Portland State University

Transcription

1 CS 410/510 Sensor Networks Portland State University Lecture 7 Energy Conservation and Harvesting 2/9/2009 Nirupama Bulusu 1

2 Source Acknowledgements Wei Ye and John Heidemann USC Information Sciences Institute Deborah Estrin Aman Kansal Mani Srivastava 2/9/2009 Nirupama Bulusu 2

3 Energy Conservation 2/9/2009 Nirupama Bulusu 3

4 Characteristics of a Sensor Network A special wireless ad hoc network Large number of nodes Battery powered Topology and density change Nodes for a common task In-network data processing Sensor-net applications Sensor-triggered bursty traffic Can often tolerate some delay Scalability & Self-configuration Energy efficiency Adaptivity Fairness not important Message-level Latency Adaptivity Trade for energy Speed of a moving object places a bound on network reaction time 2/9/2009 Nirupama Bulusu 4

5 Network-level Opportunities for Energy Conservation Radio Transmission Power Control Medium Access Control (MAC) Topology-control Routing 2/9/2009 Nirupama Bulusu 5

6 Radio Transmission Power Control Why adjust transmission power? Guarantee network connectivity Control network density/encourage spatial reuse Minimize transmission power => reduced energy consumption (also due to reduced contention) 2/9/2009 Nirupama Bulusu 6

7 Example Let R = 3r, energy consumption inversely proportional to d 2 Cost of transmitting a-d = 3 (a-b-c-d) R a b c d r 2/9/2009 Nirupama Bulusu 7

8 MAC and Its Classification Medium Access Control (MAC) When and how nodes access the shared channel Classification of MAC protocols Scheduled protocols Schedule nodes onto different sub-channels Examples: TDMA, FDMA, CDMA Contention-based protocols Nodes compete in probabilistic coordination Examples: ALOHA (pure & slotted), CSMA 2/9/2009 Nirupama Bulusu 8

9 MAC Attributes Collision avoidance Basic task of a MAC protocol Energy efficiency Scalability and adaptability Network size, node density and topology change Channel utilization Latency Throughput Fairness Primary Secondary 2/9/2009 Nirupama Bulusu 9

10 Energy Efficiency in MAC Design Energy is primary concern in sensor networks What causes energy waste? Collisions Control packet overhead Overhearing unnecessary traffic Long idle time bursty traffic in sensor-net apps Dominant factor Idle listening consumes % of the power for receiving (Stemm97, Kasten) 2/9/2009 Nirupama Bulusu 10

11 Scheduled Protocols Time Division Multiple Access (TDMA) Advantages No collisions Energy efficient easily support low duty cycles Disadvantages Poor scalability and adaptability Difficult to accommodate node changes Difficult to handle inter-cluster communication Requires strict time synchronization 2/9/2009 Nirupama Bulusu 11

12 Scheduled Protocols Polling A master plus one or more slaves (star topology) The master node decides which slave can send by polling the corresponding slave Only direct communication between the master and a slave A special TDMA without pre-assigned slots Examples IEEE infrastructure mode (CFP) Bluetooth piconets 2/9/2009 Nirupama Bulusu 12

13 Scheduled Protocols Self-Organization by Sohrabi and Pottie Have a pool of independent channels Frequency band or spreading code Potential interfering links select different channels Talk to neighbors in different time slots Sleep in unscheduled time slots Looks like TDMA, but actually FDMA or CDMA Any pair of two nodes can talk at the same time Low bandwidth utilization /9/2009 Nirupama Bulusu 13

14 Scheduled Protocols Bluetooth Target for wireless personal area network (WPAN) Short range, moderate bandwidth, low latency IEEE (MAC + PHY) is based on Bluetooth Nodes are clustered into piconet Each piconet has a master and up to 7 slaves scalability problem The master polls each slave for transmission Frequency-hopping CDMA between clusters Multiple connected piconets form a scatternet Different to handle inter-cluster communications 2/9/2009 Nirupama Bulusu 14

15 Scheduled Protocols Bluetooth (Cont.) How about Bluetooth radio with sensor networks? Scalability is a big problem Lack of multi-hop support No commercial Bluetooth radio supports scatternet so far Use two radios expensive and energy inefficient A node temporarily leave one piconet and joins another high overhead and long delay Connection maintenance is expensive even with a low-duty-cycle mode (Leopold et al.) 2/9/2009 Nirupama Bulusu 15

16 Scheduled Protocols LEACH: Low-Energy Adaptive Clustering Hierarchy by Heinzelman, et al. Similar to Bluetooth CDMA between clusters TDMA within each cluster Static TDMA frame Cluster head rotation Node only talks to cluster head Only cluster head talks to base station (long dist.) The same scalability problem 2/9/2009 Nirupama Bulusu 16

17 Contention-Based Protocols Contention-based protocols CSMA Carrier Sense Multiple Access Listening before transmitting Not enough for multi-hop networks (collision at receiver) a b c Hidden terminal: a is hidden from c s carrier sense CSMA/CA (CA stands for Collision Avoidance) RTS/CTS handshake before send data Other nodes (e.g. node c) backoff 2/9/2009 Nirupama Bulusu 17

18 Contention-Based Protocols Contention-based protocols (contd.) MACA Multiple Access w/ Collision Avoidance Add duration field in RTS/CTS informing other node about their backoff time MACAW improved over MACA RTS/CTS/DATA/ACK Fast error recovery at link layer IEEE Distributed Coordination Function (DCF) Largely based on MACAW 2/9/2009 Nirupama Bulusu 18

19 Contention-Based Protocols IEEE DCF: ad hoc mode Virtual and physical carrier sense (CS) Network allocation vector (NAV), duration field Binary exponential backoff RTS/CTS/DATA/ACK for unicast packets Broadcast packets are directly sent after CS Fragmentation support RTS/CTS reserve time for first (fragment + ACK) First (fragment + ACK) reserve time for second Give up transmission when error happens 2/9/2009 Nirupama Bulusu 19

20 Contention-Based Protocols Tx rate control by Woo and Culler Based on a special network setup A base station tries to collect data equally from all sensors in the network CSMA + adaptive rate control Promote fair bandwidth allocation to all sensors Nodes close to the base station forward more traffic, and have less chances to send their own data Helps in congestion avoidance 2/9/2009 Nirupama Bulusu 20

21 Scheduled vs. Contention Protocols Scheduled Protocols Contention Protocols Collisions No Yes Energy efficiency Scalability and adaptation Multi-hop communication Time synchronization Good Bad Difficult Strict Bad Good Easy Loose or not required 2/9/2009 Nirupama Bulusu 21

22 Energy Efficiency in Contention- Based Protocols Contention-based protocols need to work hard in all directions for energy savings Reduce idle listening support low duty cycle Better collision avoidance Reduce control overhead Avoid unnecessary overhearing 2/9/2009 Nirupama Bulusu 22

23 Energy-Efficient MAC Design PAMAS: Power Aware Multi-Access with Signalling by Singh and Raghavendra Improve energy efficiency from MACA Avoid overhearing by putting node into sleep Use separate control and data channels RTS, CTS, busy tone to avoid collision Probe packets to find neighbors transmission time Increased hardware complexity Two channels need to work simultaneously, meaning two radio systems. 2/9/2009 Nirupama Bulusu 23

24 Energy-Efficient MAC Design Piconet by Bennett, Clarke, et al. Not the same piconet in Bluetooth Low duty-cycle operation energy efficient Sleep for 30s, beacon, and listen for a while Sending node needs to listen for receiver s beacon first, then CSMA before sending data May wait for long time before sending 2/9/2009 Nirupama Bulusu 24

25 Energy-Efficient MAC Design Asynchronous sleeping by Tseng, et al. Extend PS mode to Multi-hops Nodes do not synchronize with each other Designed 3 sleep patterns ensure nodes listen intervals overlap, example: Periodically fully-awake interval: similar to S-MAC Problem on broadcast wake up each neighbor 2/9/2009 Nirupama Bulusu 25

26 Energy-Efficient MAC Design ZigBee Industry standard through application profiles running over IEEE radios Target applications are sensors networks, interactive toys, smart badges, remote controls, and home automation 2/9/2009 Nirupama Bulusu 26

27 Energy-Efficient MAC Design ZigBee (Cont.) Three devices specified Network Coordinator Full Function Device (FFD) Can talk to any device, more computing power Reduced Function Device (RFD) Can only talk to a FFD, simple for energy conservation CSMA/CA with optional ACKs on data packets Optional beacons with superframes Optional guaranteed time slots (GTS), which supports contention-free access 2/9/2009 Nirupama Bulusu 27

28 Energy-Efficient MAC Design ZigBee (Cont.) Low power, low rate (250kbps) at physical layer MAC layer supports low duty cycle operation Target node life time > 1 year 2/9/2009 Nirupama Bulusu 28

29 Case Study: S-MAC S-MAC by Ye, Heidemann and Estrin Tradeoffs Latency Fairness Major components in S-MAC Periodic listen and sleep Collision avoidance Overhearing avoidance Message passing Energy 2/9/2009 Nirupama Bulusu 29

30 Coordinated Sleeping Problem: Idle listening consumes significant energy! Solution: Periodic listen and sleep sleep listen listen sleep Turn off radio when sleeping Reduce duty cycle to ~ 10% (120ms on/1.2s off) Latency Energy 2/9/2009 Nirupama Bulusu 30

31 Coordinated Sleeping Schedules can differ Node 1 Node 2 listen sleep listen sleep listen sleep listen sleep Prefer neighboring nodes have same schedule easy broadcast & low control overhead Schedule 1 Schedule 2 Border nodes: two schedules or broadcast twice 2/9/2009 Nirupama Bulusu 31

32 Coordinated Sleeping Schedule Synchronization New node tries to follow an existing schedule Remember neighbors schedules to know when to send to them Each node broadcasts its schedule every few periods of sleeping and listening Re-sync when receiving a schedule update Periodic neighbor discovery Keep awake in a full sync interval over long periods 2/9/2009 Nirupama Bulusu 32

33 Coordinated Sleeping Adaptive listening Reduce multi-hop latency due to periodic sleep Wake up for a short period of time at end of each transmission RTS CTS CTS listen listen t1 t2 listen Reduce latency by at least half 2/9/2009 Nirupama Bulusu 33

34 Collision Avoidance S-MAC is based on contention Similar to IEEE ad hoc mode (DCF) Physical and virtual carrier sense Randomized backoff time RTS/CTS for hidden terminal problem RTS/CTS/DATA/ACK sequence 2/9/2009 Nirupama Bulusu 34

35 Overhearing Avoidance Problem: Receive packets destined for others Solution: Sleep when neighbors talk Basic idea from PAMAS (Singh, Raghavendra 1998) But we only use in-channel signaling Who should sleep? All immediate neighbors of sender and receiver How long to sleep? The duration field in each packet informs other nodes the sleep interval 2/9/2009 Nirupama Bulusu 35

36 Message Passing Problem: Sensor net in-network processing requires entire message Solution: Don t interleave different messages Long message is fragmented & sent in burst RTS/CTS reserve medium for entire message Fragment-level error recovery ACK extend Tx time and re-transmit immediately Fairness Energy Msg-level latency Other nodes sleep for whole message time 2/9/2009 Nirupama Bulusu 36

37 Implementation Platform Mica Motes (UC Berkeley) 8-bit CPU at 4MHz, 128KB flash, 4KB RAM 20Kbps radio at 433MHz TinyOS: event-driven Configurable S-MAC options Low duty cycle with adaptive listen Low duty cycle without adaptive listen Fully active mode (no periodic sleeping) 2/9/2009 Nirupama Bulusu 37

38 Experiments: Two-hop network Topology, measured energy consumption on source nodes Source 1 Source 2 Sink 1 Sink 2 Energy consumption (mj) S-MAC consumes much less energy than like protocol w/o sleeping At heavy load, overhearing avoidance is the major factor in energy savings At light load, periodic sleeping plays the key role Average energy consumption in the source nodes like protocol without sleep S-MAC w/o adaptive listen Overhearing avoidance Message inter-arrival period (second) 2/9/2009 Nirupama Bulusu 38

39 Energy Consumption over Multi-Hops Ten-hop linear network at different traffic load 3 S-MAC configurations At light traffic load, periodic sleeping has significant energy savings over fully active mode Adaptive listen saves more at heavy load by reducing latency Energy consumption (J) Energy consumption at different traffic load No sleep cycles 10% duty cycle without adaptive listen 10% duty cycle with adaptive listen Message inter-arrival period (S) 2/9/2009 Nirupama Bulusu 39

40 Latency as Hops Increase 12 Adaptive listen significantly reduces latency causes by periodic sleeping Latency under lowest traffic load 12 Latency under highest traffic load Average message latency (S) % duty cycle without adaptive listen 10% duty cycle with adaptive listen Average message latency (S) % duty cycle without adaptive listen 10% duty cycle with adaptive listen No sleep cycles Number of hops No sleep cycles Number of hops 2/9/2009 Nirupama Bulusu 40

41 Throughput as Hops Increase Adaptive listen significantly increases throughput Using less time to pass the same amount of data Effective data throughput (Byte/S) Effective data throughput under highest traffic load No sleep cycles 10% duty cycle with adaptive listen 10% duty cycle without adaptive listen Number of hops 2/9/2009 Nirupama Bulusu 41

42 Combined Energy and Throughput Periodic sleeping provides excellent 3 performance at 2.5 light traffic load No sleep cycles 2 With adaptive listening, S-MAC 1.5 achieves about the 1 10% duty cycle without same performance adaptive listen 0.5 as no-sleep mode 0 at heavy load Energy-time product per byte (J*S/byte) Energy-time cost on passing 1-byte data from source to sink 10% duty cycle with adaptive listen Message inter-arrival period (S) 2/9/2009 Nirupama Bulusu 42

43 #3: Topology Control Between MAC and routing Turn off as many nodes as possible Leave only enough on to keep a connected topology Ensures data can transit through network Topology control vs. MAC Operate at much coarser timescales Cycle radios on the order of minutes rather than seconds 2/9/2009 Nirupama Bulusu 43

44 Examples Geography-based Use physical location to infer network coverage. Divide physical area into grids, select one node per grid. Topology-based Directly measure network connectivity Select node in topology if two of its neighbors cannot talk to each other Energy savings depend on network density Node mobility 2/9/2009 Nirupama Bulusu 44

45 #4: Energy-efficient Routing Minimize energy cost per packet Balance energy consumption in the network. More in next lecture 2/9/2009 Nirupama Bulusu 45

46 Conclusion Energy conservation active area of research Current work Transmission power control MAC protocol design Topology control Routing 2/9/2009 Nirupama Bulusu 46

47 Energy Harvesting 2/9/2009 Nirupama Bulusu 47

48 Sources Chapter 9: Energy Harvesting Aman Kansal and Mani Srivastava Sensor-coordinated actuation for energy harvesting Mohammed Rahimi et al 2/9/2009 Nirupama Bulusu 48

49 Energy harvesting Batteries are too big Batteries do not last forever Methods exist to extract energy from the environment Thermoelectric (DARPA, JPL, Caltech) Micro-hydraulic transducer (DARPA, MIT) Solar cells Bio-fuel (University of Bristol) 2/9/2009 Nirupama Bulusu 49

50 Managing Harvested Energy It is different from battery energy Supply varies with time Need to adapt performance Supply varies with space Different nodes get different energy Need load sharing Supply is repetitive (does not die out) Opportunity to last forever Efficiency Concerns Match load to maximize transfer Supply direct when possible, instead of through battery 2/9/2009 Nirupama Bulusu 50

51 Example Which route to choose? source destination 2/9/2009 Nirupama Bulusu 51

52 Managing Harvested Energy Key issues Use available energy most efficiently Estimate achievable performance level Harvesting Technology Harvesting Circuit Scheduler Deployment Specific Choice Buffering (battery or ultra-capacitor) Consumption Arbitration Tracking availability Performance Scaling Network-wide task scheduling 2/9/2009 Nirupama Bulusu 52

53 Harvesting Circuit System Block Diagram Harvesting device Energy Tracker Consumption Arbiter Recharging circuit E Sub-module Power switching Sensor Node Energy storage Device data Harvesting Aware Power management 2/9/2009 Nirupama Bulusu 53

54 Scheduler Design Problem Existing Approach Kansal s Approach Environmental energy supply is variable Recharge battery and let the load use energy as desired Track environmental availability Supply is independent of demand Battery makes up for discrepancy, node dies when out of battery Scale performance; match availability; last forever 2/9/2009 Nirupama Bulusu 54

55 Source Characterization What is available from the variable environmental source? Leaky bucket like model for bursty energy supply E(t) is a (ρ,σ1,α2) source if for all T: E(t) integral over 0 to T is >= (ρt σ1) <= (ρt + σ2) 2/9/2009 Nirupama Bulusu 55

56 Interesting Questions Given the source parameters: What is the achievable application throughput or latency? Can the system last eternally at required performance level? What additional resources are required, if not? Considering efficiencies of batteries and other power modes, how should the tasks be scheduled? 2/9/2009 Nirupama Bulusu 56

57 Harvesting Theory Theorem: If a system is powered by a (ρ,σ1,σ2) source has energy capacity >= (σ1+ σ2) Operates at a constant power level ρ Then 1. It utilizes the energy source fully 2. Can survive forever 2/9/2009 Nirupama Bulusu 57

58 Performance control Learn energy environment parameters Predict Sustainable Performance level ρ = xp max + (1 x)p sleep Adapt Performance Sleep and active modes Dynamic voltage scaling Radio range control Sub-module power switching 2/9/2009 Nirupama Bulusu 58

59 Multi-server Harvesting How can a distributed system manage the harvested energy to maximize performance of the system as a whole? Energy resources vary across nodes Task-load differs at different nodes Some workload is shareable while some is not Consider one energy intensive task: routing data Determine environmental energy aware communication strategy 2/9/2009 Nirupama Bulusu 59

60 Routing Options Optimal routing is impractical Nodes share state information and coordinate performance adaptation actions Nodes adapt performance locally and routing protocol operates over sleepy nodes 2/9/2009 Nirupama Bulusu 60

61 Practical Networking Method Routing for an event monitoring sensor network Single sink (base station), multiple sources (node monitoring events) Must report event when it occurs; otherwise no data Measure energy and calculate duty-cycle locally Duty cycle determines latency of data relaying Sensor Sleep Timer Event detected Timer expired Sensor Node Snooze: Processor And radio sleeping 2/9/2009 Nirupama Bulusu 61

62 Communication with Sleep Mode Node can wake up if it has data to send How does a sleeping node receive data? 2/9/2009 Nirupama Bulusu 62

63 Routing Tree INIT Base station ACK Protocol Base station sends INIT Receiver sends ACK and forwards INIT Reverse path set up to base station Possibly shortest; but not necessarily lowest latency 2/9/2009 Nirupama Bulusu 63

64 Network-wide Performance Estimate network-wide latency constraint with observed environmental resource Central control over network latency is impractical Sending all latencies to base station reduces scalability Use in-network processing to compute path latency Receive path latencies from children Forward highest plus own latency 2/9/2009 Nirupama Bulusu 64

65 Conclusions Harvesting technologies can enable long system lifetime Proof-of-concept system and algorithms to exploit environmental energy demonstrated Methods are needed to measure and characterize energy sources Battery characterization is not sufficient Distributed methods are required to optimally adapt global performance Schedule tasks appropriately in space and time to enhance performance 2/9/2009 Nirupama Bulusu 65