Wireless Sensor Networks: Application Driver for Low Power Systems Deborah Estrin Laboratory for Embedded Collaborative Systems (LECS) UCLA Computer Science Department http://lecs.cs.ucla.edu destrin@cs.ucla.edu 1 Applications Scientific: eco-physiology, biocomplexity mapping Infrastructure: contaminant flow monitoring (and modeling) www.jamesreserve.edu Engineering: monitoring (and modeling) structures 2 Common Vision • Embed numerous distributed devices to monitor and interact with physical world • Exploit spatially and temporally dense, in situ, sensing and actuation • Network these devices so that they can coordinate to perform higher-level tasks • Requires robust distributed systems of hundreds or thousands of devices 3 Challenges • Tight coupling to the physical world and embedded in unattended “control systems” – Different from traditional Internet, PDA, Mobility applications that interface primarily and directly with human users • Untethered, small form-factor, nodes present stringent energy constraints – Living with small, finite, energy source is different from traditional fixed but reusable resources such as BW, CPU, Storage • Communications is primary consumer of energy in this environment – R4 drop off dictates exploiting localized communication and innetwork processing whenever possible 4 New Design Themes • Long-lived systems that can be untethered and unattended – Low-duty cycle operation with bounded latency – Exploit redundancy – Tiered architectures (mix of form/energy factors) • Self configuring systems that can be deployed ad hoc – Measure and adapt to unpredictable environment – Exploit spatial diversity and density of sensor/actuator nodes 5 Approach • Leverage data processing inside the network – Exploit computation near data to reduce communication • Achieve desired global behavior with adaptive localized algorithms (i.e., do not rely on global interaction or information) – Dynamic, messy (hard to model), environments preclude pre-configured behavior – Cant afford to extract dynamic state information needed for centralized control or even Internetstyle distributed control 6 Why cant we simply adapt Internet protocols and “end to end” architecture? • Internet routes data using IP Addresses in Packets and Lookup tables in routers – Humans get data by “naming data” to a search engine – Many levels of indirection between name and IP address – Works well for the Internet, and for support of Person-to-Person communication • Embedded, energy-constrained (un-tethered, small-form-factor), unattended systems cant tolerate communication overhead of indirection 7 Techniques for building long-lived • Exploiting redundancy – Adaptive Self-Configuration – Supporting low-duty cycle operation • Exploiting heterogeneity 8 Exploiting Redundancy: Goal • To extend system lifetime • We may be able to deploy 100 times as many nodes in environments where we can’t increase the battery capacity by factor of 100 • To overcome environmental limitations (obstructions) • Non line of site conditions, Variable sensor coupling • To achieve good coverage with ad-hoc deployment • When deployment or operational conditions cant be controlled precisely 9 Exploiting Redundancy example • Efficient, multi-hop topology formation goal: exploit redundancy provided by high density to extend system lifetime while providing communication and sensing coverage. – If too many sensors active at the same time, increase energy consumption and competition for communication resources. – If too few nodes active, then lack of communication and/or sensing coverage. – Central control/configuration requires too much communication – Nodes should self-configure to find the right trade-off – Ultimately should adapt based on desired “fidelity” 10 Adaptive Fidelity Examples • ASCENT – Node measures number of neighbors and packet loss to determine participation, duty cycle, and/or power level. – Ratio of energy used by Active case (all nodes turn on) to energy used by ASCENT • GAF – Uses Geographic information to infer which nodes might be redundant with one another for the purposes of routing • Open question: Can we apply Adaptive Fidelity etmore generally? 11 • Ratio of energy used by the Active case (all nodes turn on) to the energy used by ASCENT • ASCENT provides significant energy savings over the Active case 12 Robustness and Scalability through Adaptation • Adaptive mechanisms increase complexity but enable self-configuration for robustness and scalability • Self calibration to adapt to variations in sensor response and placement • Adjust duty cycle and transmit range as a function of node density and measured range (adaptive fidelity) – Balance increased system life-time with increased resolution • Challenge: develop and evaluate localized adaptive algorithms • We hope adaptive functions will go beyond “connectivity”…e.g., tracking 13 Supporting low duty cycle operation • S-MAC – A MAC designed for wireless sensor networks by increasing and facilitating sleep time and reducing overhearing and contention energy expenditure • Triggering and tracking – Use lower-power modalities, devices, to trigger higher power ones – Use active devices to trigger sleeping devices to increase fidelity – Paging channels 14 Supporting low duty cycle operation • S-MAC – – – – Message passing Periodic listen/sleep Avoid overhearing Energy Measurement • On motes and TinyOS • Two-hop network with 2 sources and 2 sinks • Under different traffic load 15 Adaptive Tracking Example • Network nodes close to tracked event (or with good data on the event) enter fully active state; other nodes dormant/low duty cycle • Sentry nodes active; wake up dormant nodes when necessary. • Wakeup wavefront precedes phenomenon being tracked. • Information driven diffusion (Zhao, Reich, et.al.): node propagates expression for evaluating best next node(s) in wavefront based on information utility and cost • Requires: – low power operating mode with wake up/paging channel – definition of a wakeup wavefront using localized algorithms – time synchronization 16 Low Duty Cycle Time Synchronization • Pulse synchronization creates locality of synchronized nodes, quickly and efficiently – “External” node generates pulse. Synchronizing nodes compare reception times. – NTP good at correcting frequency – Local pulse good at correcting phase – Use combination 17 18 Exploiting Heterogeneity: Tiered Architecture • Technological advances will never prevent the need to make tradeoffs • Nodes will need to be faster or more energy-efficient, smaller or more capable or more durable. • Tiered platform consisting of a heterogeneous collection of hardware. – Larger, faster, and more expensive hardware (sensors) – Smaller, cheaper, and more limited nodes (tags and motes) 19 Tiered Architecture • Discover and exploit asymmetry wherever possible – Base stations for aggregating resources; motes for access to physical phenomena – Variable power, distance radios • E.g., nodes in ASCENT can adapt by reducing their radio range, using less energy and reducing channel contention. – Multiple modalities • E.g., localization with RF, Acoustics, and Imaging 20 Can we eliminate the finite nature of the energy source? • Batteries will provide 1J/mm3 (Pister) • When available, solar has a lot (the most) to offer in recharging (Pister) • Other possibilities: Charging the batteries on fields of sensors by driving through them ? 21 Current Research Areas • Constructs for “in network” distributed processing – system organized around naming data, not nodes • Programming large collections of distributed elements • Localized algorithms that achieve systemwide properties • Time and location synchronization – energy-efficient techniques for associating time and spatial coordinates with data to support collaborative processing • Experimental infrastructure 22 Current COTS Infrastructure PC-104+ (off-the-shelf) UCB Mote (Culler/Hill/Pister) Software • Directed Diffusion • TinyOS (UCB/Culler) • Measurement, Simulation 23 Embedded, Everywhere A Research Agenda for Networked Systems of Embedded Computers • Fall 2001: Computer Science and Telecommunications Board report (late September) • Recommends major areas of research needed to achieve robust, scalable EmNets – predictability, adaptive selfconfiguration, monitoring & system health, computational models, network geometry, interoperability, social and policy issues • Substantive recommendations to DARPA, NIST, & NSF For more information, see www.cstb.org or contact lmillett@nas.edu 24 Future Directions • Proposed Center for Embedded Networked Sensing (CENS) – Develop technology architecture, software, components in the context of driving application prototypes • • • • Habitat monitoring/Biocomplexity mapping Seismic activity and structure response Contaminant flow monitoring Grades 7-12 science curricula innovations 25 Acknowledgments • Funders – DARPA SenseIT and NEST Programs http://www.darpa.mil/ito/research/sensit – NSF Special Projects – Cisco, Intel • Collaborators – UCLA LECS students: Bien, Bulusu, Busek, Braginsky, Bychkovskiy, Cerpa, Elson, Ganesan, Girod, Greenstein, Perelyubskiy, Scoellhammer, Yu http:/lecs.cs.ucla.edu/ – USC-ISI Collaborators Govindan, Heidemann, Intanago, Silva, Wei, Zhao http://www.isi.edu/scadds – UCB Intel Lab: Culler, et.al. 26