AETOS is able to reflect graph properties that optimize applications of self-organizing tree overlay networks. For example, AETOS shorts the nodes of a tree by positioning the most robust ones close to the root and the least robust nodes as leaves. In this way, the impact of node failures, security attacks or excessive load is minimized as the size of the affected branch is smaller compared to an unsorted tree. Furthermore, AETOS is able to reflect constraints of nodes in the tree topology such as the node degree (number of connected children). The bandwidth of nodes in a multicasting tree is an example of such application constraints. AETOS is based on such abstract criteria related with the graph properties of trees to meet a wide range of application requirements.
Self-organization in AETOS is designed in three incremental and reconfigurable levels:
- Gossiping discovery: Nodes disseminate and collect information in an epidemic fashion. This creates a dynamic and random network that keeps all the nodes interconnected.
- Clustering: Candidate parents and children, ranked according to one or more application criteria, are sorted and selected based on an adaptation strategy, e.g. highly ranked parents are preferred over low ranked ones.
- Tree building and maintenance: Nodes negotiate with the selected candidate parents and children for the establishment of their in-between connection. The decision-making is based on the application requirements and results in different tree topologies, e.g. short trees consisting of a small number of tree levels with a large number of nodes or long trees consisting of a large number of tree levels with a small number of nodes.
This incremental self-organization methodology benefits from its abstraction, customization and modularity compared to other methodologies that are more integrated with specific applications.
AETOS was part of my PhD studies and resulted in several publications.
- Evangelos Pournaras, Martijn Warnier and Frances M.T. Brazier, Adaptive Self-organization in Distributed Tree Topologies, International Journal of Distributed Systems and Technologies, Vol. 5, Nr. 3, pp. 24-57, 2014 © IGI Global ( Best paper award )
- Evangelos Pournaras, Multi-level Reconfigurable Self-organization in Overlay Services, PhD Thesis, Delft University of Technology, March 2013 (Chapter 4)
- Evangelos Pournaras, Martijn Warnier and Frances M.T. Brazier, Self-optimized Tree Overlays using Proximity-driven Self-organized Agents, book chapter in Complex Intelligent Systems and Their Applications, in the Springer Book Series ”SPRINGER OPTIMIZATION AND ITS APPLICATIONS”, Vol. 41, Chap. 7, pp. 137-161, June 2010. © Springer-Verlag
- Evangelos Pournaras, Martijn Warnier and Frances M.T. Brazier, Adaptation Strategies for Self-management of Tree Overlay Networks, in the proceedings of the International Conference on Grid Computing-Grid 2010, pages 401-409, Brussels, Belgium, October 2010. © IEEE/ACM [Presentation]
- Evangelos Pournaras, Martijn Warnier and Frances M.T. Brazier, Adaptive Agent-based Self-organization for Robust Hierarchical Topologies, in the proceedings of the International Conference on Adaptive and Intelligent Systems-ICAIS 2009, pages 69-76, Klagenfurt, Austria, September 2009. © IEEE [Presentation]
- Evangelos Pournaras, Martijn Warnier and Frances M.T. Brazier, Adaptive Agent-based Self-organization for Robust Hierarchical Topologies, in the proceedings of the 7th European Workshop on Multi-agent Systems-EUMAS 2009, Ayia Napa, Cyprus, December 2009 (compressed contribution of an earlier paper) [Presentation]
- Evangelos Pournaras, A Survey and a Proposed Approach on Robust Tree Overlays, for the ASCI course on Fundamentals and Design of Distributed Systems, Vrije Universiteit Amsterdam, The Netherlands, June 2009 [Presentation]