Directed Diffusion for Wireless Sensor Networking C Intanagonwiwat

Directed Diffusion for Wireless Sensor Networking C. Intanagonwiwat, R. Govindan, D. Estrin, John Heidemann, and Fabio Silva ACM/IEEE 2003 발표자 : 문동규

Contents w Introduction w Directed Diffusion n n n Simplified schematic for Directed Diffusion Data Naming Interest & Gradient Interest Propagation Data Propagation Reinforcement w Simulations w Conclusion

Introduction (1/2) w Wireless sensor networks n n Sensing devices with communication capability Event monitoring l Enemy detection, aircraft interiors, large industrial plants A sensor field Sources Even t w Data-centric communication n n Data is named by attribute-value Different form IP-style communication l End-to-end delivery service Directed Diffusion Sink Node

Introduction (2/2) w Data-centric communication (cont. ) n n Human operator’s query (task) is diffused Sensors begin collecting information about query Information returns along the reverse path Intermediate nodes aggregate the data l Combing reports from sensors w Challenges n n Scalability Energy efficiency Robustness / Fault tolerance in outdoor areas Efficient routing

Data Naming w Content based naming n n n Task are named: Attribute – value pair Selecting naming scheme important No globally unique ID for nodes: only locally unique Request: Interest Reply: Data type = four-legged animal interval = 20 ms interval = 1 s duration = 10 seconds rect = [-100, 200, 200] timestamp = 01: 20: 40 expires. At = 01: 30: 40

Interest & Gradient w Interest describes a task needed to be done by the sensor network n n n Interests are injected into the network at sink. Sink broadcasts the interest. Interval specifies an event data rate. Initially, requested interval much larger than needed. Node maintains an interest cache. w Interest entry also maintains gradients. n n Specifies a data rate and a direction (neighbor) Data flows from the source to the sink along the gradient

Interest Propagation w Flooding w Constrained or Directional flooding based on location. w Directional Propagation based on previously cached data. Gradient Source Interest Sink

Data Propagation w Reinforcement to single path delivery. w Multipath delivery with probabilistic forwarding. w Multipath delivery with selective quality along different paths. Gradient Source Data Sink

Reinforcement w Reinforce one of the neighbor after receiving initial data. n n n Neighbor(s) from whom new events received. Neighbor who’s consistently performing better than others. Neighbor from whom most events received. Gradient Source Data Reinforcement Sink

Negative Reinforcement(1/2) w Explicitly degrade the path by re-sending interest with lower data rate. w Time out Gradient Source Data Reinforcement Sink

Negative Reinforcement(2/2) w Using negative reinforcement n n Path Truncation Loop removal l l For resource saving B negative reinforces D, D negative reinforces E, …

Performance Evaluation (1/7) w w w Simulator: ns-2 Network Size: 50 -250 Nodes Transmission Range: 40 m Constant Density: 1. 95 x 10 -3 nodes/m 2 (9. 8 nodes in radius( MAC: Modified Contention-based MAC Transceiver Energy Model: mimics a “sensor radio” n 660 m. W in transmission, 395 m. W in reception, and 35 m. W in idle w Comparison with n n Flooding Omniscient multicast

Performance Evaluation (2/7) w Average dissipated energy Average Dissipated Energy (Joules/Node/Received Event) 0. 018 0. 016 Flooding 0. 014 0. 012 0. 01 0. 008 Omniscient Multicast 0. 006 0. 004 Diffusion Due to the data-aggregation Nodes suppress duplicate location estimates 0. 002 0 0 50 100 150 200 Network Size 250 300

Performance Evaluation (3/7) w Average delay Average Delay (secs) 0. 35 0. 3 Flooding 0. 25 0. 2 0. 15 0. 1 0 Omniscient Multicast Diffusion 0. 05 0 50 100 150 200 250 Uncongested sensor 300 network Reinforcement rules find the low delay path Network Size

Performance Evaluation (4/7) w Impact of dynamics (Distinct event delivery ratio)

Performance Evaluation (5/7) w Impact of negative reinforcement Diffusion Without Negative Reinforcement Diffusion With Negative Reinforcement Prune off higher latency path

Performance Evaluation (6/7) w Impact of duplicate suppression Diffusion Without Suppression Diffusion With Suppression Negative reinforcement Suppress identical data sent

Performance Evaluation (7/7) • High idle radio power

Conclusion w Application-level data dissemination has the potential to improve energy efficiency significantly n n n Data-centric dissemination Reinforcement based adaptation of paths Duplicate suppression and aggregation w MAC for sensor networks needs to be designed carefully.