Dynamic Forwarding over Tree-on-DAG for Scalable Data Aggregation in Sensor Networks

1. Dynamic Forwarding over Tree-on-DAG for Scalable Data Aggregation in Sensor Networks High-Speed Networking Lab. Dept. of CSIE, Fu-Jen Catholic University Adviser: Jenn Wei Lin Speaker: Tzung-Lin Yu

2. Outline • Abstract • Introduction • Related Work • DAA (Data Aware Anycast) • Dynamic Forwarding over ToD • One dimentional • Two dimentional • Performance Evaluation • Conclusion • Reference

3. I. Abstract • Computing and maintaining network structures for efficient data aggregation incurs high overheadfor dynamic events where the set of nodes sensing an event changes with time. • We propose Tree on DAG (ToD), a semistructured approach that uses Dynamic Forwarding to support to make the network efficient aggregation in large-scale networks.

4. Direct Communication Sensor node Base station

5. LEACH

6. II. Introduction • Sensor networks data aggregation can often reduce the communication cost by eliminating redundancy. • Various structured approaches for data aggregation have been proposed for data gathering applications and event-based applications. fixed structures cannot efficiently aggregate data change the structure dynamically incur high maintenance overhead

7. II. Introduction • We propose an efficient and scalable data aggregation mechanism that can achieve early aggregationwithout incurring overhead of constructing a structure.

8. III. Related Work • DAA (Data Aware Anycast) • Structureless protocol • Packets have to be transmitted to the same node at the same time to be aggregated • Spatial convergence:forward to the nodes that have packets (by radio) • Temporal convergence:using Randomized Waiting to increase the chance Disadvantage: • if no neighbor has packets for aggregation  to sink (high cost) or return to DAA • Does not guaranteethe aggregation of all packets when the network grows.

9. III. Related Work • Fixed tree structures • Long stretch problem:Packets from adjacent nodes have to be forwarded many hops away before aggregation. Sink EventSource

10. IV. Dynamic Forwarding over ToD • When no further aggregation can be achieved, we forward packets on ToD instead of forwarding to the sink (DAA). • We propose a dynamic forwarding mechanism over ToD to avoid the long stretch problem of fixed structure.

11. IV. Dynamic Forwarding over ToD • ToD in One-Dimentional Network • basic of Two-Dimentional • Network is divided into cells • cell: a square (length = ) > Max-diameter of an eventcan span

12. S-aggregator F-aggregator IV. Dynamic Forwarding over ToD • F&S-Tree, Cell, Cluster, and Aggregator • Each F-aggregator creates a shortest path (SPT) to the sink. • F-clusters’ size is large enough to cover the cells an event can span.

13. IV. Dynamic Forwarding over ToD • For all sets of nearby cells that can be triggered by an event, either they will be in the same F-cluster, or they will be the same S-cluster. to avoid the long stretch problem

14. IV. Dynamic Forwarding over ToD Dynamic Forwarding • using the DAA to aggregate • no further aggregation:forwarding their packets to their F-aggregators(i) event in single F-cluster using F-tree forward the packets to the sink ex: cell A, B(ii) event in multiple F-clusterselect the S-aggregator for further aggregation ex: cell C, D [Property1]For any two adjacent nodes in ToD in 1D network, their packets willbe aggregated either at an F-aggregator or S-aggregator.

15. IV. Dynamic Forwarding over ToD ToD in Two-Dimentional Network • ToD in 1D works because an event always in the same F-cluster or S-cluster. • In 2D scenarios, if an event spans multiple F-clusters, each F-aggregator may have multiple choices of S-aggregators. • Assumption:1. size of a grid cell > maximum size of an event2. event is contiguous3. Dynamic Forwarding requires each F-aggregator knows the location of S-aggregator.

16. IV. Dynamic Forwarding over ToD • 5 × 5 F-clusters F-Cluster cell F-aggregator S-Cluster F-Cluster

17. IV. Dynamic Forwarding over ToD • an event spans cells in the (1) same F-cluster  aggregated at the F-aggregator(2) multiple F-clusters • four basic scenarios Generate Packets( Source) Corresponding S-cluster

18. IV. Dynamic Forwarding over ToD (2) multiple F-clusterspackets originate from (i) 3 or 4 cells in the same F-clusterno other nodes in other F-clusters have packets forward to the sink F -cluster F -cluster

19. F -cluster F -cluster F -cluster (ii) 1 or 2 cellspossible that other F-clusters also have packet(a) 1 cell: in the same S-cluster F-aggregator forward packets to S-aggregator (b) 2 cells: must be in different S-cluster  3 cases : F -cluster S -cluster

20. IV. Dynamic Forwarding over ToD • Case 2: X-Cluster  2 cell, Y-Cluster  1 cell • to guarantee that the packets can meet at least at one S-aggregator  two F-aggregatorsselect one S-aggregator (closer to the sink) be the 2nd S-aggregators • S-aggregator only forwards packets to the 2nd S-aggregators if the packets it received come from two cells in one F-cluster • 2nd S-aggregator wait longer than the 1st S-aggregator

21. IV. Dynamic Forwarding over ToD [Property2]any two adjacent nodes in ToD, their packet will be aggregated at the F-aggregator, at the 1stS-aggregator, or at the 2nd S-aggregator • even if the size of event is not known, this approach can work and efficiently than DAA

22. IV. Dynamic Forwarding over ToD • F&S-clusters select an aggregator • nodes play this role in turn distribute the energy consumption • nodes can elect themselves • Frequency of updating can be low. • Using a hash functionto hash current time to a node.select the kth node be the aggregator

23. IV. Dynamic Forwarding over ToD choose a Aggregating Cluster • to Simplify the cluster-head selection process • Choose an F-cluster, called Aggregating Cluster, for each S-cluster. (closet to the sink) • Use the F-aggregator of Aggregating Cluster as the S-cluster’s S-aggregator. • The common aggregator for both the shaded F-cluster and S-cluster Select the Aggregating Cluster aggregator instead of S-Cluster S F Aggregating Cluster aggregator (Use the F-aggregator as the S-aggregator)

24. V. Performance Evaluation • Kansei sensor testbed • Nodes: 105 Mica2-based motes • 7×15 grid network with 3-ft spacing • Each mote is hooked onto a Stargate • Stargate: • a 32-bit hardware device from CrossBow running Linux • be connected to the server using wired Ethernet • program motes, send messages and signals to motes through Stargate • Radio signal: using default transmission power covers most nodes • Limit nodes only to receive packets from two-grid neighboring nodes neighbors : Each node has a maximum of 12 neighbors • Event size: not limit • Generated an Event report: Node is triggered by an event (store in a report queue) • Both the application layer and Anycast MAC layer can access the report queue • Divide the network into 2 F-clusters in ToD • The smallest cell to have only 9 sensor nodes • do not consider energy of consumption on idle listening

25. V. Performance Evaluation • Protocol • Dynamic Forwarding over (ToD) • DAA: structureless approach • SPT: node send packets to the sink through the SPTimmediately after sensing an event • SPT-D (SPT with Fixed Delay ): SPTwith delay according to their height • Normalized number of transmissions (NNT)= Number of transmissions in the entire network ÷ useful information from sources to the sink

26. V. Performance Evaluation • NNT vs. Event Size • fixed event location • Diameter: 12 ~ 36 ft • Node 2~6 grid-hops of the event will be triggered • Sink : at one corner Performance: • Size , ToD  (more chancesto aggregated) • Size ,SPT-D (long stretch problem) • fixed structured affects performance significantly

27. V. Performance Evaluation • NNT vs. Maximum Delay • Delay: 0 ~ 8 sec. • All node generate one packet every 10 sec. Performance: • SPT-D (structured-based) heavily depends on the delay

28. V. Performance Evaluation • Large-Scale Simulation • NS2 simulator • ToD, DAA, SPT and OPT • 2000 × 1200 m grid network with 35-m node separation • 1938 nodes • Transmission range of nodes: >50m • Event moves: random waypoint mobility model at speed of 10 m/s for 400 seconds.Event Size: 400m in diameter • OPT (Optimal Aggregation Tree): • nodes forward their packets on the aggregation tree • aggregation tree:rooted at center of the event • Nodes know where to forward packet to and how long to wait • change when event moves • not considerate the construction overhead

29. NNT vs. Event Size NTT vs. Event Size V. Performance Evaluation closer to the source Aggregate packets early OPT: best performance, but overhead not considered Total Unit of Useful Info. Received by Sink vs. Event Size

30. V. Performance Evaluation NNT vs. Distance to the Sink NTT vs. Distance to the Sink The lower the better(ratio of aggregation is high) Number of Packet Received at the sink per event vs. Distance to the Sink

31. VI. Conclusion • We proposed a semistructured approach. • Dynamic Forwarding on ToD to avoid the long stretchproblem in fixed structured and eliminates the overhead of constructing and maintaining dynamic structures.

32. VII. Reference • Dynamic Forwarding over Tree-on-DAG for Scalable Data Aggregation in Sensor NetworksFan, Kai-Wei; Liu, Sha; Sinha, Prasun;Mobile Computing, IEEE Transactions onVolume 7,  Issue 10,  Oct. 2008 Page(s):1271 - 1284 Digital Object Identifier 10.1109/TMC.2008.55