Sensor Networks and the Future of Networked Computation

Size: px

Start display at page:

Download "Sensor Networks and the Future of Networked Computation"

Carmella Robertson
5 years ago
Views:

1 Sensor Networks and the Future of Networked Computation James Aspnes Yale University February 16th, 2006

2 Why wireless sensor networks? Rationale Classical networks The present Question: If a tree falls in the forest, how do we hear it?

3 Why wireless sensor networks? Rationale Classical networks The present Question: If a tree falls in the forest, how do we hear it? Answer: nail a sensor to every tree.

4 Rationale Classical networks The present The central dogma of sensor networks Distributed sensing is necessary to detect rare, localized events. Abundant sensors must be cheap. Short-hop radio is the obvious communication mechanism. The more computation we can do in the network itself, the less communication we need.

5 Classical networks Rationale Classical networks The present Classical networks are made of big, expensive devices: Routers. Wires (or equivalent fixed point-to-point connections). Power cables running to the routers. Network administrators standing next to the routers.

6 Rationale Classical networks The present are made of small, cheap devices: Sensors. Short-distance radio broadcast. Weak power sources: batteries, solar cells, RF antennas. No network administrators!

7 Rationale Classical networks The present are made of small, cheap devices: Sensors. Short-distance radio broadcast. Weak power sources: batteries, solar cells, RF antennas. No network administrators! Every feature of sensor networks is likely to become typical of most networked computing devices.

8 Where we are evolving from Rationale Classical networks The present Existing sensor network applications. Weather stations. Networks of strain gauges on ships. Ecological sensor networks. B-52-launched tank detectors. Active RFID chips. Mostly used for inventory tracking. Many are writable. Some already can record temperature extremes or sudden drops. New hardware technologies like proximity interconnect and chip-based networks.

9 Where we are evolving to Rationale Classical networks The present Sensors/RFID chips that talk to each other. Computers without wires. Cheap sensors everywhere. Long view: smart molecules?

10 Classic network problems revisited New problems New models Classic network problems revisited Consensus, leader election, clock synchronization with severe resource constraints. New geographical routing algorithms that require no external configuration and respect sensors limitations. Internet-scale problems on much smaller scales.

11 Classic network problems revisited New problems New models Data aggregation The central problem in sensor networks. Still an active area of research.

12 Classic network problems revisited New problems New models Localization How do we know where our sensors are? Triangulation/trilateration. NP-hard in the worst case. Good algorithms for dense networks. Error propagation still needs work.

13 Classic network problems revisited New problems New models Mobility Controlled mobility creates a need for planning. Uncontrolled mobility requires tolerance of a rapidly-changing network structure (or maybe no consistent network structure). Details: see previous talk.

14 Classic network problems revisited New problems New models Security How do we keep dumb sensors from being hijacked or misused? Controversy over adding RFID tags to US passports exemplifies issues of control. Most RFID and sensor network applications rely on physical distance to limit access. Such reliance may not be reasonable as sensors become ubiquitous.

15 Classic network problems revisited New problems New models New models Population protocols model collections of very weak devices using networks of finite-state automata. Unstructured case is now very well understood. Bounded-degree networks are in principle equivalent to LINSPACE Turing machines. Speedy algorithms are still needed. Can we build continuous models that are good approximations to discrete physical systems? Battery life, geographical constraints, and sensor unreliability need to be incorporated into models at a fundamental level.

16 There are many more bacteria than people (even by weight). The same economics suggests we can expect many more tiny computers than big ones. are the vanguard of this revolution

17 There are many more bacteria than people (even by weight). The same economics suggests we can expect many more tiny computers than big ones. are the vanguard of this revolution.

Engr 1202 ECE. Clean Room Project

Engr 1202 ECE. Clean Room Project Engr 1202 ECE Clean Room Project Dilbert the engineer gets special recognition September 2005 2014 Version does not even have my name! AC vs. DC Circuits DC and AC devices in everyday life DC Devices