1. 2004/2/10 2004/2/10Jenchi Minimum-Energy Asynchronous Dissemination to Mobile Sinks in Wireless Sensor Networks ACM Conference on Embedded Networked Sensor Systems (SenSys 2003), Los Angeles, USA, Nov 2003 Hyung Seok Kim, Tarek F. Abdelzaher, and Wook Hyun Kwon

2. 2004/2/11Jenchi Outlines Introduction Introduction Assumptions and Basic Model Assumptions and Basic Model SEAD SEAD Related Works Related Works Evaluation Evaluation Conclusion Conclusion

3. 2004/2/11Jenchi Introduction Sinks Sinks –gathering the sensor readings –may be mobile PDAs carried by users or may be static access points –have different service requirements  desired data refresh rate  end-to-end delay between the source and the sink

4. 2004/2/11Jenchi Introduction (cont.) Energy is identified as the most crucial resource in sensor networks Energy is identified as the most crucial resource in sensor networks –The energy consumption of each sensor node  the cost of transmitting and receiving messages When sinks are mobile in sensor networks When sinks are mobile in sensor networks –Communication consists of three main parts  Building the dissemination tree (d-tree)  Disseminating data  Maintaining linkage to mobile sinks

5. 2004/2/11Jenchi Introduction (cont.) A Scalable Energy-efficient Asynchronous Dissemination protocol (SEAD) A Scalable Energy-efficient Asynchronous Dissemination protocol (SEAD) –A distributed self-organizing protocol that saves communication energy –Sinks are mobile –Does not use mobile sinks as intermediate members of the tree  This precludes frequent changes of the dissemination path due to sink mobility –A stationary sensor node takes the mobile sink ’ s place for building an optimal dissemination tree

6. 2004/2/11Jenchi Assumptions and Basic Model Basic Assumptions Basic Assumptions –Each sensor node is aware of its own geographic location  Not require GPS at every node –After having been deployed, sensor nodes remain stationary at their initial locations –The sensor nodes are homogeneous –Sensor nodes communicate with sinks by delivering data across multiple hops

7. 2004/2/11Jenchi Assumptions and Basic Model (cont.) Overview of the algorithm Overview of the algorithm –One source generates the sensory update traffic possibly on behalf of a group of local sensors –The data updates are disseminated along a tree to the mobile sinks in an asynchronous manner –Each branch of the tree may have its own update rate  depending on the desired refresh rate of the downstream observers

8. 2004/2/11Jenchi Assumptions and Basic Model (cont.)

9. 2004/2/11Jenchi Assumptions and Basic Model (cont.) Overview of the algorithm (cont.) Overview of the algorithm (cont.) –U : the average update rate of the source –U m : a minimum update rate,  to detect failures or packet loss in the sensor network  If a source has no new sensor readings –It disseminates idle messages along the tree at rate U m –If a node in the tree receives no messages including idle messages from the source for a period longer than 1/ U m  The node contacts its parent  If its parent has failed and gives no response, –The node asks for a new parent by sending an error message to the source of the d-tree

10. 2004/2/11Jenchi Assumptions and Basic Model (cont.) When a mobile sink wants to join the d-tree When a mobile sink wants to join the d-tree –It selects one of its neighboring sensor nodes to send a join query to the source of the tree –the selected node is called the sink ’ s access node The access node The access node –is used to represent the moving sink when the optimal d-tree is built –Amortize the overhead in the presence of mobility

11. 2004/2/11Jenchi Assumptions and Basic Model (cont.) The tree delivers data to the fixed access node. In turn, the access node delivers the data to the sink without exporting the sink ’ s location information to the rest of the tree The tree delivers data to the fixed access node. In turn, the access node delivers the data to the sink without exporting the sink ’ s location information to the rest of the tree The tree is updated only when the access node changes The tree is updated only when the access node changes As the sink moves, no new access node is chosen until the hop count between the access node and the sink exceeds a threshold As the sink moves, no new access node is chosen until the hop count between the access node and the sink exceeds a threshold –Trade-off between path delay and energy spent on reconstructing the tree

12. 2004/2/11Jenchi Assumptions and Basic Model (cont.) Source data is replicated at selected nodes between the source and sinks Source data is replicated at selected nodes between the source and sinks We define a replica as a sensor node that stores a copy of the source data We define a replica as a sensor node that stores a copy of the source data –It temporarily stores the latest data incoming from the source and asynchronous disseminate it to others along the tree

13. 2004/2/11Jenchi Assumptions and Basic Model (cont.)

14. 2004/2/11Jenchi Assumptions and Basic Model (cont.) Definition Definition –V _ the sensor nodes in the sensor network –B _ sinks indexed by n=1,2,…,N and m=1,2,…,M –A_ A={A 1,A 2,…,A M } –A_ A={A 1,A 2,…,A M } ⊂ V be the set of M access nodes {B 1,B 2,…,B M } which request for mobile sinks B={B 1,B 2,…,B M } which request data from the source at refresh rates data from the source at refresh rates R={R 1,R 2,…,R M } R={R 1,R 2,…,R M } –Energy_cost(a,b) –Energy_cost(a,b) ∝ d(a,b)P ab  P ab — packet sending rate  d(a,b) —distance between a and b

15. 2004/2/11Jenchi SEAD Subscription Query Subscription Query –Sink directs a join query to source via its access node Gate replica search Gate replica search –A gate replica is determined, which serves as the grafting point from which a branch to the new access point is extended Replica placement Replica placement – –locally readjusts the tree in the neighborhood of the gate replica to further reduce communication energy D-tree management D-tree management – –The constructed tree is managed to accommodate mobile sinks or defective regions such as a group of congested or failed nodes

16. 2004/2/11Jenchi SEAD – Subscription Query Mobile sinks beacon periodically to determine their neighbors Mobile sinks beacon periodically to determine their neighbors A mobile sink B i selects the nearest of its adjacent nodes as the access node A i A mobile sink B i selects the nearest of its adjacent nodes as the access node A i –B i sends a join query to a source via A i –The join query message contains the location of the A i – and the sink ’ s desired updated rate R i The access node directly sends the join query to the source via the underlying routing protocol The access node directly sends the join query to the source via the underlying routing protocol

17. 2004/2/11Jenchi SEAD — Gate Replica Search Each node n in the tree has a set C(n) of children Each node n in the tree has a set C(n) of children It maintains a downstream rate Q n c for each child cC(n) It maintains a downstream rate Q n c for each child c ∈ C(n) The algorithm starts when a source receives a query indicating a sink ’ s desired refresh rate, R i The algorithm starts when a source receives a query indicating a sink ’ s desired refresh rate, R i Each level of replica r, including the source receiving that message, runs a recursive search as follows Each level of replica r, including the source receiving that message, runs a recursive search as follows –If Q n c < R i,then change Q n c value to R i

18. 2004/2/11Jenchi SEAD — Gate Replica Search (cont.) When a replica r is connected to the access node A i, the additional cost, K(r), is calculated as When a replica r is connected to the access node A i, the additional cost, K(r), is calculated as

19. 2004/2/11Jenchi SEAD — Gate Replica Search (cont.) Consider the incremental cost, Consider the incremental cost, K(r)-K(c) where r is c ’ s parent K(r)-K(c) where r is c ’ s parent Which relates the cost at node r to that at its child c (4) Which relates the cost at node r to that at its child c (4)

20. 2004/2/11Jenchi SEAD — Gate Replica Search (cont.)

21. 2004/2/11Jenchi SEAD — Replica Placement Locally adjusts the tree around the gate replica to produce an optimal d-tree There are two ways to connect the access node to the gate replica There are two ways to connect the access node to the gate replica –Non-replica mode  Connect it as a child of the gate replica  Adds no replicas to the tree –Junction mode  Create a child for the gate replica to feed the access node and some of the gate replica ’ s original children  The new child replica is called a junction replica

22. 2004/2/11Jenchi SEAD — Replica Placement (cont.) The replica placement phase The replica placement phase –compares the non-replica mode cost U non_replica to a junction replica mode cost U jreplica –Selects the better option so that the access node joins the tree in a way that minimizes the energy cost

23. 2004/2/11Jenchi SEAD — Replica Placement (cont.) The gate replica g U non_replica The gate replica g calculates the cost of the non- replica mode U non_replica (c) for each child c ∈ C(g)

24. 2004/2/11Jenchi SEAD — Replica Placement (cont.) U jreplica The gate replica g finds the neighbor node n among its adjacent nodes within a singe hop range. Then it calculates the energy cost U jreplica (c) for each child c

25. 2004/2/11Jenchi SEAD — Replica Placement (cont.) After getting both U non_replica (c) and U jreplica (c), the gate replica finds one of its children cC(g) to maximize After getting both U non_replica (c) and U jreplica (c), the gate replica finds one of its children c ∈ C(g) to maximize

26. 2004/2/11Jenchi SEAD — Replica Placement (cont.)

27. 2004/2/11Jenchi SEAD — D-tree management Sink mobility Sink mobility

28. 2004/2/11Jenchi SEAD — D-tree management (cont.) Sink mobility Sink mobility

29. 2004/2/11Jenchi SEAD — D-tree management (cont.) Sink mobility Sink mobility

30. 2004/2/11Jenchi SEAD — D-tree management (cont.) Sink mobility Sink mobility

31. 2004/2/11Jenchi SEAD — D-tree management (cont.) Sink mobility Sink mobility

32. 2004/2/11Jenchi SEAD — D-tree management (cont.) Leaving D-Tree Leaving D-Tree –A sink B i leaves the d-tree by sending a leave message to its access node –The access node A i requests its parent to delete A i from its list of children and stop forwarding data to it

33. 2004/2/11Jenchi SEAD — D-tree management (cont.) System Lifetime System Lifetime Avoiding defective points or areas Avoiding defective points or areas Complexity and resource requirements Complexity and resource requirements

34. 2004/2/11Jenchi Evaluation Simulation ： NS-2 A popular sensor node prototype ： MICA2 Power is supplied via disposable AA batteries. – –About 0.080W for transmitting and About 0.025W for receiving. The transceiver has a 250m radio range at 433 MHz, which is the case with the radio transceiver of a MICA2 mote. The sensor network consists of (300 ≤ N ≤ 500) sensor nodes in a 2000m × 2000m grid or 3-5 nodes per 200m × 200m use the two-ray ground model as the radio propagation model and an omni-directional antenna having unity gain in the simulation Each query packet is 36 bytes long and the data packet has 64 bytes The default number of sinks is 8. Three different sources generate different data at an average interval of 6 seconds The desired update rates are generated at random The energy consumption is measured in terms of Joules per node. The default mobile sink speed is set to 10 m/sec (i.e., the fastest human speed)

35. 2004/2/11Jenchi Evaluation (cont.) Compare SEAD to – –Directed Diffusion (DD) [15]   A data-centric communication paradigm specifically designed for sensor networks   Subscribers use flooding to spread interests to the sensor network – –TTDD [25]   Multicast protocol   Exploits local flooding within a local cell of a grid which sources build proactively   Each source disseminates data along the nodes on the grid line to the sinks – –ADMR [16]   Mobile ad hoc network

36. 2004/2/11Jenchi Evaluation — Asynchronous dissemination (cont.) Energy consumed for data packets Energy consumed for data packets

37. 2004/2/11Jenchi Evaluation — Asynchronous dissemination (cont.) Average delay Average delay

38. 2004/2/11Jenchi Evaluation — Sink mobility Energy consumption for the number of sinks Energy consumption for the number of sinks

39. 2004/2/11Jenchi Evaluation — Sink mobility (cont.) Energy consumption with different sink speeds Energy consumption with different sink speeds

40. 2004/2/11Jenchi Evaluation — Sink mobility (cont.) Remaining energy distribution of the sensor nodes Remaining energy distribution of the sensor nodes

41. 2004/2/11Jenchi Evaluation — Impact of the node density Energy per node with different sensor node density Energy per node with different sensor node density

42. 2004/2/11Jenchi Evaluation — Impact of the node density (cont.) Average delay with different sensor node density Average delay with different sensor node density

43. 2004/2/11Jenchi Evaluation — End-to-end delay of SEAD Average delay with different number of sinks Average delay with different number of sinks

44. 2004/2/11Jenchi Evaluation — End-to-end delay of SEAD (cont.) Average delay with different sink speeds Average delay with different sink speeds

45. 2004/2/11Jenchi Conclusion SEAD SEAD –Saves energy consumption in both building the d-tree and maintaining linkage to mobile sinks –Strikes a balance between end-to-end delay and power consumption that favors power savings over delay minimization –Is insensitive to changes in node density