1. Christian Frank, Kay Römer ETH Zurich Algorithms for Generic Role Assignment in Wireless Sensor Networks

2. ACM SenSys 2005, Nov 2-4 2 read_sensor() send_msg() get_pos() read_sensor() send_msg() get_pos() The Gap

3. ACM SenSys 2005, Nov 2-4 3 Generic Role Assignment  Enables automatic assignment of –Special functions/roles to nodes in the network –Using programmer-specified rules for assignment  Rules are based on local and neighborhood properties Coverage on off Clustering chslavegw Aggregation srcsinkagg

4. ACM SenSys 2005, Nov 2-4 4 Role Specifications Use Case / Architecture Property Directory RA Algorithm Gateway App. Network Sensor Node battery = 80% pos = (12.3, 3.4) role = ON … Simulation & Evaluation

5. ACM SenSys 2005, Nov 2-4 5 Clustering Appl. CLUSTERHEAD :: { battery >= 60% && count(1 hop) { role == CLUSTERHEAD } == 0 } GATEWAY :: { chs == retrieve(1 hop, 2) { role == CLUSTERHEAD } && count(2 hops) { role == GATEWAY && chs == super.chs } == 0 } SLAVE :: else  p == retrieve(scope, num) { pred } –At least num nodes in scope must fulfil pred –Bind p to ids of matching nodes  count(scope) { pred }: –Counts nodes matching pred within scope

6. ACM SenSys 2005, Nov 2-4 6 Role Specifications Use Case / Architecture Property Directory RA Algorithm Gateway App. Network Sensor Node Simulation & Evaluation

7. ACM SenSys 2005, Nov 2-4 7  Local cache table on each node –Contains local and remote properties  Algorithm consists of three procedures: 1)Initialize cache table 2) Propagate properties to neighbors 3) Choose role according to local table –On change of local table: Reschedule 2) and 3)  Iteration through a set of roles  Notify applications on stable role Distributed Algorithm Clustering

8. ACM SenSys 2005, Nov 2-4 8 1)Initialization 2) Property Propagation –broadcast all rows x with dist < max and dirty == true –set x.dirty to false 3) Local Rule Evaluation Distributed Algorithm 0 Dist 1 Max trueundef.roleA DirtyValueKeySrc A BC ON :: { count(1 hop) { role == ON } == 0 } OFF :: else ON OFF undef. false11ONroleA 0 Dist 1 Max falseundef.roleB DirtyValueKeySrc false11OFFroleC trueOFF A BC

9. ACM SenSys 2005, Nov 2-4 9 Probabilistic Initialization  Improve convergence –Chose initial role smartly  Approach: –Estimate probability p r for each role –Draw role r with probability p r –Estimation can be done offline using static information Specification Node degree estimate  Extension: –Combine estimate p r and known/certain information  Later on “repair” inconsistent role assignments –Using standard cache table approach

10. ACM SenSys 2005, Nov 2-4 10 Probabilistic Initialization ON:: count(1) { role == ON } <= lim  Given: –Specification –Estimated n nodes within scope –Initial role probabilities  Compute role probabilities from spec. –Consider above example, probability that: k out of n nodes are ON k nodes less/eq. lim are ON –Assumption: Symmetric probabilities  System of equations – solved offline using fixpoint iteration

11. ACM SenSys 2005, Nov 2-4 11  Additionally given –Est. role probabilities (last slide) –Roles of some nodes in scope  Compute role probability given known roles  Make use of initial specification flood Wave Initialization Sink Y nodes are known and ON ON:: count(1) { role == ON } <= lim X unheard-of nodes yet expected in scope

12. ACM SenSys 2005, Nov 2-4 12 Role Specifications Use Case / Architecture Property Directory RA Algorithm Gateway App. Network Sensor Node Simulation & Evaluation

13. ACM SenSys 2005, Nov 2-4 13  Simulation tool –Discrete event simulator based on JIST/SWANS –Visualization / specification frontend –Specification compiler  Network model –Based on CC1000 parameters –Simple CSMA approach, only broadcast is used –Intentionately, no measures to improve reliability  Initial prototype on real nodes –Supports subset of specification (count operators) Implementation

14. ACM SenSys 2005, Nov 2-4 14  Simulated three specifications –Coverage / clustering / aggregation  Studied algorithms –Basic caching algorithm –Basic + probabilistic initialization –Basic + wave-based initialization  Examined… –Overhead, while varying nodes in same area –Convergence, while varying nodes in same area no. of role changes until a stable role is reached –Robustness, while varying an additional ratio of lost messages –Proportionality, while varying the maximum scope of the specification Evaluation

15. ACM SenSys 2005, Nov 2-4 15 Convergence  Metric –Num. of role changes (except 1 st prob. choice)  Coverage results –No further reconfiguration after wave  Clustering results –Probabilistic does not improve

16. ACM SenSys 2005, Nov 2-4 16 Limitations / Discussion  Some specifications may not terminate –Support user to detect non-terminating specifications –Simulation tool used for testing –Protect deployed network by limiting role changes  Cannot describe every algorithm –Focus on ease-of-use for application domain experts –Extensible by using app.-specific procedures  Efficiency –Effort proportional to “difficulty” of specification –Comparable to “specific” implementations

17. ACM SenSys 2005, Nov 2-4 17 Conclusion/Outlook  First generic role assignment tool –System service for WSN configuration problems –Used to formulate a variety of network configuration heuristics –Rapid prototyping  System properties –Proportional effort –Efficient probabilistic initialization  Future work –TinyOS implementation –More flexible scope definitions –Adaptation for more heterogeneous networks

18. Thank you!