1. Observability by Design Principles of Observable Sensor Networks J. Beutel F. Mattern K. Römer L. Thiele

2. 2 Goal: Make Sensor Networks Observable  Minimal or no interference  Systematic and methodology-based  Facilitate deployment  Improve reliability

3. 3 Example  Project Permasense

4. 4 Based on Phase II Project  WSN Diagnosis using Deployment Support Network –overhearing message traffic Temporary deployment support network sensor network  Infer basic network state –Approximate neighborhood –Routing topology  Many symptoms are detectable by passive inspection  E.g., basic failure indicators –Dead nodes –Network partitions Observe network state on deployment site

5. 5 Some Relevant Publications 2007  M. Ringwald, K. Römer, A. Vitaletti: Passive Inspection of Sensor Networks, Proc. 3rd IEEE Conf. Distributed Computing in Sensor Systems (DCOSS 2007).  M. Ringwald, K. Römer: Deployment of Sensor Networks: Problems and Passive Inspection, Proc. 5th Work- shop on Intelligent Solutions in Embedded Systems (WISES 2007).  M. Dyer, J. Beutel, T. Kalt, P. Oehen, L. Thiele, K. Martin, P. Blum: Deployment Support Network, Proc. 4th European Conf. Wireless Sensor Networks (EWSN 2007).  M. Wöhrle, C. Plessl, J. Beutel and L. Thiele: Increasing the Reliability of Wireless Sensor Networks with a Unit Testing Framework, Proc. 4th Workshop on Embedded Networked Sensors (EmNets 2007).

6. 6 Limitations of Passive Inspection  Some important information cannot be inferred, e.g. –Battery level, clock drift, power consumption –Did a certain node receive a broadcast message?  Message losses difficult to detect  Only approximations for certain state info (e.g., neighborhood of a node)  Primary reasons for failures symptoms often unclear  Ad hoc “solutions” –Add specific message –Add sequence numbers, sender ID,… to packets –Protocol changes (e.g., acks)  Systematic / generic?  Specification of (applica- tion specific) observable state / behavior?  Automatic generation? Observable Sensor Networks Design and Implementation of Design Methodology for

7. 7 A Framework for Observability Application Observability requirements Trace analysis Back annotation Observable state Energy/ resource budget Protocols include extra bits of state info in headers Generator Executable code

8. 8 Research Agenda  Metrics –Observability, energy cost of observability  Networking protocols and programming models –Add a small amount of information to message headers –Specify observable system state –Time-triggered paradigm  Trace analysis –Analyze gathered execution traces to infer state, failure indicators and causes  Back annotation –Correlate failure causes with programming models  Test in application context (e.g., Permasense)

9. Observability by Design Principles of Observable Sensor Networks J. Beutel F. Mattern K. Römer L. Thiele

11. 11 Related Work  Stanford (MNet) –Share goal of design for observability –Different approach (in-band), less comprehensive (protocol design)  UCLA (Sympathy) –Collection of metrics and detection of events to identify failure states  Virginia (Clairvoyant) –Node-level debugger for sensor nodes –Heavy impact on sensor node behavior, not practical for true networking problems

12. 12 Core Research Challenges  Genericity –Provide reusable abstractions and frameworks rather than ad hoc solutions  Incomplete information –Allow reliable observation of system state despite incomplete/noisy information  Resource constraints –Manage tradeoff between observability and energy/resource consumption

13. 13 Summary  Observability is a critical property of sensor networks to ensure their correct operation –Establish observability as a main design goal  Advance state of the art through provision of –A comprehensive and generic observability framework –Networking protocols, programming models, trace analysis, back annotation  Contribute to the use of large-scale sensor networks for real-world applications  We expect mutual benefit through cooperation with application projects

14. 14 Resources  3 PhD –Programming models, networking protocols –Trace analysis –Back annotation

15. 15 Reliability?  Sensor networks are fragile –Harsh environment, scare resources effort 100% reliability (“working correctly”) Failure reasons often unclear trial-and-error