Prediction Assisted Single-copy Routing in Underwater Delay Tolerant Networks

1. Prediction Assisted Single-copy Routing in Underwater Delay Tolerant Networks ZhengGuo, Bing Wang and Jun-Hong Cui Computer Science & Engineering Department, University of Connecticut, Storrs, CT, 06269 IEEE Globecom 2010

2. Outline • Introduction • Network Model • Aggressive Chronological Projected Graph (ACPG) • Prediction Assisted Single-Copy Routing (PASR) • Performance Evaluation • Conclusion

3. Introduction • Many routing protocols have been proposed to deal with the lack of contemporaneous end-to-end paths in delay tolerant networks (DTNs). • Due to node mobility and sparse node deployment, UWSNs can be treated as DTNs. • Limited bandwidth • High power consumption • Mobility patterns

4. Introduction • UWSNs are extremely resource stringent since acoustic communication. • Furthermore, the mobility patterns in an UWSN can vary dramatically over time depending on the environment. • These two characteristics render existing multi-copy based DTN routing protocols unsuitable for UWSNs. • Waste Energy

5. Goal • Proposed a generic scheme, prediction assisted single-copy routing (PASR), for UWSNs. • PASR can be instantiated to efficient single-copy routing protocols under different mobility models.

6. Network Model • Consider a data collection underwater sensor network, which consists of M layers. Current Link Each Sensor： Surface Sink Buffer Layer 1 Layer 2 Battery Sensor Water Currents …… Layer M Super Source

7. Aggressive Chronological Projected Graph (ACPG) • Proposed a greedy algorithm to capture the network mobility properties and the common characteristics of near optimal routes. • ACPG compresses the evolving network topology and connectivity to a single graph 𝐺(𝑉,𝐸) chronologically, efficiently finds routes from the graph in a slot by slot manner.

8. Aggressive Chronological Projected Graph • Construction G(V, E) • Vertex vij∈ V is the jth node in the ith layer. • Edge (vij, vkl) ∈Erepresents the connection between these two nodes during a certain time slot 𝑡. the capacity of that connection (u,v,t,C) (v21, v12) vd Layer 1 v11 v12 v13 v14 v Layer 2 v21 v22 v23 v24 u Layer 3 v31 v32 v33 v34 vs

9. Aggressive Chronological Projected Graph A. Construction G(V, E) • Vertex vij∈ V is the jth node in the ith layer. • Edge (vij, vkl) ∈Erepresents the connection between these two nodes during a certain time slot 𝑡. • Node v ∈ V can be in two status: inactive or active. • Node v • Uv：The upstream node. • Iv(i) ：The maximum number of packets that can be transmitted or received during the ith slot. • Cv(i)：The available storage in the ith slot • Pv ：The residual power for transmissions vd Layer 1 v11 v12 v13 v14 Layer 2 v21 v22 v23 v24 Layer 3 v31 v32 v33 v34 vs

10. Aggressive Chronological Projected Graph B. Operations of ACPG • The operations of ACPG at each slot t, t > 0, include two routines: • 1.Edge projection • during which connections in a time slot are projected to G as edges • 2.Routes reservation and graph update • during which routes are discovered and G is updated. (u,v,t,C) vd v11 v12 v13 v14 Layer 1 v Layer 2 v21 v22 v23 v24 u Layer 3 v31 v32 v33 v34 vs (v32 ,v23,2,5 ) (v32 ,v23,5,4 )

11. Aggressive Chronological Projected Graph • An example of projected graph for 12 nodes in 3 layers. Inactive node Active node (vd←v12← v22← v31← vs) Capacity：5 Packets Delay：8 slots vd (9,10) (9,5) Layer 1 v11 v12 v13 v14 (7,2) (7,7) (vd←v12←v22← v23← v33← vs) Capacity：2 Packets Delay：7 slots (6,6) (5,3) Layer 2 v21 v22 v23 v24 (3,5)/(7,9) (7,4) (4,6)/(8,5) Layer 3 v31 v32 v33 v34 (1,5) (2,7) vs

12. Performance of ACPG 600m 600m Layer 1 Layer 2 • Evaluate the performance of ACPG by comparing it with optimal solutions from integer linear programming (ILP). • Each sensor has • Buffer size=30 packets • Transmission range=100 m • 3rd layer generating packets from the 500th second to the 1000th second with the rate of one packet per second. Layer 3 90m

13. Performance of ACPG • Comparison of ILP and ACPG.

14. Prediction assisted single-copy routing (PASR) A. How to works PASR? • Propose prediction assisted single-copy routing (PASR), that utilizes ACPG in a training period to capture the characteristics of the mobility pattern, and provide guidance on route selection. • Historical information • Guidance from ACPG • Predict the future

15. Historical information • If the mobility pattern is stable for a long time, the history can tell the future. The most widely used historical information includes. • Recent trajectory • Average contact duration • Average inter-contact duration • Last contact time • Contact frequency

16. Guidance from ACPG • The following properties of routes and node contacts, which are closely related to the underlying mobility pattern, can be captured by ACPG: • Geographic preference • Contact periodicity • Contact probability

17. Predict the future • After ACPG characterizes the mobility pattern, it suggests what historical information can be used for prediction.

18. Prediction assisted single-copy routing (PASR) 800m 800m Layer 1 Layer 2 B. Instantiating PASR • Consider three mobility models in an underwater sensor network. • UWSN in Regular Currents • UWSN in Currents with Randomness • UWSN in Irregular Currents • Each sensor has • Buffer size：100 packets • Transmission range：50m • Transmission rate：50 packet/s • Power capacity：300 to 30 • Slot duration：10s Layer 3 40m

19. UWSN in Regular Currents • 1)Guidance from ACPG: Focus on two properties: 1.Geographic preference 2.Contact periodicity. Sink Layer 1 Layer 2 Layer 3 Super Source Geographic preference

20. UWSN in Regular Currents • 2) Protocol following ACPG: Based on the above guidance, proposed a specific PASR for this network, energy efficient history prediction assisted routing (EEHPA).

21. UWSN in Regular Currents (EEHPA) • 2) Protocol following ACPG This scheme includes two essential operations: 1. Prediction update Each node u maintains its own prediction vector (PV), which is a vector of tuples (i,v,Dv). • i is the prediction slot • v is the best relay in this slot • Dv is the expected delay through this relay to the sink 2.Per-contact forwarding decision • (1) Dv∈[Dv’ , Dv’ + δ1) • (2) Dv∈[Dv’ + δ1, Dv’ + δ2] and node v is in the upper layer. PV v u δ1,δ2 are called prediction error tolerances

22. UWSN in Currents with Randomness (EEHPA) • The randomness models the impact from environment, which may lead to estimation errors and prediction errors in real systems. • PASR can tolerant these errors to some extent since ACPG just captures the general properties of the majority of nodes, who exhibit similar mobility patterns.

23. UWSN in Irregular Currents (iEEHPA) • Assume that nodes in the first two layers will be affected by an irregular water current. • Modify EEHPA to obtain a new PASR, named iEEHPA, according to the new guidance. • Two guidance from ACPG: (1) A node in the same layer is preferred (2) Only predict for nodes in the same layer Sink Irregular water current Layer 1 Anchored node 10s Layer 2 Regular current Layer 3 Super Source

24. Performance Evaluation 800m 800m Layer 1 Layer 2 Sink • Each sensor has • Buffer size=100 packets • Transmission range=50m • Transmission rate=50 packet/s • Power capacity = 300 to 30 • Slot duration = 10s • Bottom layer randomly generate 300 packets from the 500th second with the total generation rate of one packet per second. Layer 3 40m Super Source

25. Performance Evaluation EEPA : without kinematic model FC :First Contact Epidemic : A flooding based scheme

26. Performance Evaluation

27. Performance Evaluation

28. Conclusion • Present a generic scheme prediction, assisted single-copy routing (PASR), for UWSNs. • Design online heuristic protocols by choosing appropriate historical information and forwarding criteria based on the guidance from ACPG. • Investigate an UWSN with various mobility patterns and randomness using two instantiated PASR schemes, EEHPA and iEEHPA. • Simulation results show that ACPG captures the properties of various mobility patterns and provides corresponding guidance, and the instantiated PASR schemes outperform others.

29. Thanks for your attention Happy Lunar New Year