310 likes | 510 Views
Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks. Intanagonwiwat, Govindan, Estrin USC, Information Sciences Institute, UCLA. Carl Hartung CSCI 7143: Secure Sensor Networks. Overview. Directed Diffusion Conventions and Terms Interest Propagation
E N D
Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Intanagonwiwat, Govindan, Estrin USC, Information Sciences Institute, UCLA Carl Hartung CSCI 7143: Secure Sensor Networks
Overview • Directed Diffusion • Conventions and Terms • Interest Propagation • Data Propagation • Reinforcement • Summary • Evaluation of Directed Diffusion • Impacts of node failures, etc..
Directed Diffusion • A Data Driven routing protocol • The basics: • A node (sink) broadcasts out Interests • If a node measures something of interest, send it back to interested node. • Every node thinks all neighbors are End Points • Localized repair and reinforcement • Multi-path delivery for different sinks
Naming • Task descriptions are named by Attribute-Value pairs • Query/interest: • Type=four-legged animal // detect animal location • Interval=20ms (event data rate) // send back events every 20 ms • Duration=10 seconds // for the next 10 seconds • Rect=[-100, 100, 200, 400] // from sensors within rectangle • Reply: • Type=four-legged animal // type of animal seen • Instance = elephant // instance of this type • Location = [125, 220] //location of node sensing • Intensity = 0.6 // signal amplitude measure • Confidence = 0.85 // confidence • Timestamp = 01:20:40 // event generation time
Interests and Gradients • Interests injected into network by (possibly arbitrary) node– now called sink. • Interests are cached by all nodes for time=duration, then purged • Interests are periodically refreshed by the sink. • Low initial data rate • Higher if found something of interest
Interests cont’d • Nodes cache many interests • Cached interests do not contain info about the sink – only node it received interest from • Interest entry contains possibly many gradient fields
Gradients • Contain a data rate field requested by the specified neighbor • Also contains timestamp and expiresAt • One per neighbor per Interest • Each interest can have many gradients (one per neighbor)
Interest Propagation (flooding) C A F D B G E
Interest Propagation (flooding) C A F Interests D Sink B G E
Interest Propagation (flooding) C A F Interests D Sink B G E
Interest Propagation (flooding) C A F Interests D Sink B G E
Interest Propagation (flooding) C A F Interests D Sink B G E
Data Propagation Sensed something that matched an interest C A F D Sink B G E
Data Propagation C A F D Sink B G E
Data Propagation C A F D Sink B G E
Data Propagation (ignored) C A F D Sink B G E
Data Propagation (ignored) C A F D Sink B G E
Reinforcement C A F Re-send Interest with smaller interval D Sink B G E
Reinforcement C A F Re-send Interest with smaller interval D Sink B G E
Reinforcement C A F Primary path D Sink B G E
Summary • Data-centric communication • All communication neighbor to neighbor, not end-to-end • All neighbors appear to be ‘end’ to each node • Routes are established ‘on demand’ • Message cache used to avoid loops
Analysis • Used 2 metrics to measure • Average dissipated energy • Measures the ratio of total dissipated energy per node in the network to the number of distinct events seen by sinks • Average Delay • Measures the average one-way latency observed between transmitting an event and receiving it at the sink • Simulation uses a 1.6Mbps 802.11 MAC layer
Analysis • Compared Directed Diffusion to 2 other protocols • Flooding • All events are flooded to every node in the network • Omniscient Multicast • Each source transmits events along shortest-path multicast tree to all sinks
Problems? • Interest timeouts while data is en-route to sink. • Congested network? • Can the network satisfy small event data intervals? • Multiple? • Security – nodes temporarily disabled cause data to loop? • Cache size / Timeouts
Conclusion • Directed Diffusion has potential for significant energy efficiency • Robust in dynamic sensor networks • Self Configuring • A good start • Need better evaluation