1. Adaptive Self-Configuring Sensor Network Topologies ns-2 simulation & performance analysis Zhenghua Fu Ben Greenstein Petros Zerfos

2. Sensor Networks Advances in micro-sensor and radio technology will enable deployment of sensors for a range of environmental monitoring applications Due to the low cost per node, networks of sensors may be densely distributed

3. Challenge Unattended sensor nets with limited energy often must be long-lived The network should be able to adaptively self-configure to maximize energy efficiency, while still achieving spatial coverage and robustness

4. ASCENT Adaptive Self-Configuring sEnsor Networks Topologies Answers how to form a multi-hop topology Developed by Alberto Cerpa, UCLA Protocol assumes dense distribution of nodes ASCENT leverages the redundancy imposed by high node density Each node assesses its connectivity and adapts its participation in the multi-hop network topology Network membership determined in a distributed fashion using measurements and calculations performed locally

5. What is ASCENT? It is NOT a routing or data dissemination protocol ASCENT simply decides which nodes should join the routing infrastructure In this respect, routing protocols are complementary to ASCENT

6. Why not use a central configuration node? Scaling and robustness considerations Nodes would need to communicate detailed connectivity state information to this central node

7. Network Assumptions Ad-hoc deployment Energy constraints Unattended operation under dynamics In many such contexts, it will be easier to deploy large numbers of nodes initially rather than to deploy additional nodes or energy reserves at a later date.

8. Effects of Node Density Too few nodes Larger inter-node distance Higher packet loss rate Too many nodes At best, unnecessary energy expenditure At worst, interfering nodes congest the channel Equilibrium Approximated using self-configuration

9. ASCENT Initially, only nodes A and B are alive All other nodes are passively listening, but are not part of the network A B

10. ASCENT A sends data to B Due to signal attenuation and random shadowing B detects high message loss Notion of “high” is application dependent A B

11. A B B attempts to remove this “communication hole” B requests additional nodes in the region to join the network to serve to relay messages from A to B ASCENT

12. A B Additional node(s) join the network Alternatively, while passively listening, node C determines whether it would be “helpful” to join the multi-hop routing infrastructure C ASCENT

13. Neighbor Discovery Phase A neighbor is defined as a node from which a node receives a certain percentage of messages over time The number of neighbors can greatly increase the energy consumption in contention for resources

14. Neighbor Discovery Phase Entered at time of node initialization Using local measurements, each node obtains an estimate of the number of neighbors actively transmitting messages As the neighbor count increases, so too should the neighbor’s message loss threshold

15. Join Decision Phase A node decides whether to join the multi-hop diffusion sensor network A node may temporarily join and test whether it contributes to improved connectivity The decision is based on message loss percentage and number of neighbors

16. Active Phase A node enters the Active Phase from the Join Decision Phase when it decides to join the network for a long time Starts sending routing control and data messages

17. Adaptive Phase When a node decides NOT to join the network A node has the option of either powering down for a period of time or reducing its transmission range

18. Previous ASCENT Implementation PC-104 nodes RPC Radiometrix Radio Linux LECS barebones CSMA MAC Diffusion Routing ASCENT written on top of Diffusion ASCENT uses Diffusion for routing of its control messages

19. ASCENT for NS Rewrote ASCENT for NS Placed the ASCENT code in the NS Link Layer Removed calls to Diffusion to route control packets Modified 802.11 and Link Layer code so that nodes could “play dead” Removed retransmit functionality of 802.11

20. Motivation ASCENT has been tested on only 25 nodes. With NS it can be tested on hundreds Verify that NS models the real world well Confirm that ASCENT works

21. Simulation Setup Communication within 1-hop distance 90% message loss at 1100m. One source (sending frequencies of 1p/sec and 10p/sec – CBR-like), one sink Densities of : 3, 4, 5, 10, 30, 50, 75 and 100 nodes, randomly distributed in a fixed area (800m x 800m) MAC layer: IEEE 802.11 with no retransmissions - messages sent in broadcast Propagation model : Shadowing (probabilistic)

22. Metrics Message Loss End-to-end percentage of data packets created by source that were correctly received by sink Event Delivery Ratio Percentage of packets that could have been received that actually were received (at each node) Delay (end-to-end) Overhead Percentage of packets received (at each node) that were control packets

23. Message Loss vs. Density

24. Event Delivery vs. Density

25. Delay vs. Density

26. Overhead vs. Density

27. Message Loss vs. Density

28. Event Delivery vs. Density

29. Delay vs. Density

30. Overhead vs. Density

31. Conclusions Our experiments (2Mbps), Previous experiments (13Kbps) Under-utilized channel Small vulnerable Period Propagation effects examined better than collision effects ASCENT performs well As # nodes increases (scalability) As sending frequency increases our overhead remains low