4.5 Conclusion

This chapter has presented a per component eventual leader election algorithm for dynamic networks that shows the advantages of all nodes using network topology knowledge for the choice of the leader. To this end, by exchanging messages, every node maintains a local knowledge of the communication graph of connected nodes and exploits such knowledge to elect as the leader the node having the highest closeness centrality. This leader can, therefore, spread information faster over its connected component than flooding algorithms. Considering the random waypoint and a periodic single point of interest mobility models, both versions of the Topology Aware algorithm (based on closeness and degree centralities respectively) and a flooding algorithm with a local topological election criterion, were evaluated on PeerSim [MJ09] simulator.

A performance comparison of the Topology Aware algorithm with a variant of the leader election algorithm of Vasudevan et al. [VKT04] is presented, because their work is a good example of a typical flooding algorithm and is strongly referenced in the literature [RAC08; Ing+13; KW13]. The results confirm the effectiveness of the Topology Aware algorithm and that it outperforms the latter. Both Topology Aware algorithms are more stable than Flooding Degree and the Closeness version has a shorter path to the leader, especially on large components with low movements of nodes. The latter is less sensitive to the component size and sends fewer messages than Flooding Degree. When compared to Flooding Degree, both Topology Aware versions improve the leader stability up to 82% depending on mobility models, sends half as many messages, and nodes reach the leader by 11% shorter paths. It is worth pointing out that the size of messages in Topology Aware could be reduced using compression, for example.

One limitation of the work presented in this chapter is the assumption of reliable channels. Therefore, the second algorithm presented in the next chapter considers eventually reliable communication channels, with interference, collision, and messages loss.

Table of Contents

1 Introduction
1.1 Contributions
1.1.1 Topology Aware Leader Election Algorithm for Dynamic Networks
1.1.2 Centrality-Based Eventual Leader Election in Dynamic Networks
1.2 Manuscript Organization
1.3 Publications
1.3.1 Articles in International Conferences
1.3.2 Articles in National Conferences
2 Background
2.1 Properties of Distributed Algorithms
2.2 Timing Models
2.3 Process Failures
2.4 Communication Channels
2.5 Failures of Communication Channels
2.6 Distributed Systems
2.6.1 Static Systems
2.6.2 Dynamic Systems
2.7 Centralities
2.8 Messages Dissemination
2.9 Leader Election
2.9.1 Classical Leader Election
2.9.2 Eventual Leader Election
2.10 Conclusion
3 Related Work
3.1 Classical Leader Election Algorithms
3.1.1 Static Systems
3.1.2 Dynamic Systems
3.2 Eventual Leader Election Algorithms
3.2.1 Static Systems
3.2.2 Dynamic Systems
3.3 Conclusion
4 Topology Aware Leader Election Algorithm for Dynamic Networks
4.1 System Model and Assumptions
4.1.1 Node states and failures
4.1.2 Communication graph
4.1.3 Channels
4.1.4 Membership and nodes identity
4.2 Topology Aware Leader Election Algorithm
4.2.1 Pseudo-code
4.2.2 Data structures, variables, and messages (lines 1 to 6)
4.2.3 Initialization (lines 7 to 11)
4.2.4 Periodic updates task (lines 12 to 16)
4.2.5 Connection (lines 20 to 23)
4.2.6 Disconnection (lines 24 to 27)
4.2.7 Knowledge reception (lines 28 to 38)
4.2.8 Updates reception (lines 39 to 53)
4.2.9 Pending updates (lines 54 to 65)
4.2.10 Leader election (lines 17 to 19)
4.2.11 Execution examples
4.3 Simulation Environment
4.3.1 Algorithms
4.3.2 Algorithms Settings
4.3.3 Mobility Models
4.4 Evaluation
4.4.1 Metrics
4.4.2 Instability
4.4.3 Number of messages sent per second
4.4.4 Path to the leader
4.4.5 Fault injection
4.5 Conclusion
5 Centrality-Based Eventual Leader Election in Dynamic Networks
5.1 System Model and Assumptions
5.1.1 Node states and failures
5.1.2 Communication graph
5.1.3 Channels
5.1.4 Membership and nodes identity
5.2 Centrality-Based Eventual Leader Election Algorithm
5.2.1 Pseudo-code
5.2.2 Data structures, messages, and variables (lines 1 to 4)
5.2.3 Initialization (lines 5 to 7)
5.2.4 Node connection (lines 8 to 17)
5.2.5 Node disconnection (lines 18 to 23)
5.2.6 Knowledge update (lines 24 to 34)
5.2.7 Neighbors update (lines 35 to 41)
5.2.8 Information propagation (lines 42 to 47)
5.2.9 Leader election (lines 48 to 52)
5.3 Simulation Environment
5.3.1 Algorithms Settings
5.3.2 Mobility Models
5.4 Evaluation
5.4.1 Metrics
5.4.2 Average number of messages sent per second per node
5.4.3 Average of the median path to the leader
5.4.4 Instability
5.4.5 Focusing on the 60 meters range over time
5.4.6 A comparative analysis with Topology Aware
5.5 Conclusion
6 Conclusion and Future Work
6.1 Contributions
6.2 Future Directions
A Appendix
A.1 Energy consumption per node
A.1.1 Simulation environment
A.1.2 Algorithms settings
A.1.3 Mobility Models
A.1.4 Metric
A.1.5 Performance Results