Adaptive Agent-Based Self-Organization for Robust Hierarchical Topologies

Evangelos Pournaras, Martijn Warnier, Frances M. T. Brazier
2009 2009 International Conference on Adaptive and Intelligent Systems  
Virtual organizations in large-scale distributed environments can organize their communication in a hierarchical topology (i.e., trees). However, such topologies can be unreliable as local failures have a global impact in the organization. Hierarchical topologies need to adapt continuously to changes of the underlying environment. Pro-active and re-active self-organization can make such topologies highly robust. This paper proposes AETOS, the Adaptive Epidemic Tree Overlay Service. AETOS is a
more » ... w agent-based approach for building and maintaining on-demand robust tree topologies that structure communication. Agents are pro-actively (self)-organized appropriately in a tree to minimize the effect of failures. In addition, they re-actively rewire their connections to reflect changes in the environment. The selforganization model, the control of the system and an illustrative example are discussed in this paper. * Affiliated with Delft University of Technology from the 1st of September 2009 Adaptive central control of hierarchical topologies for communication is not always an option nor a scalable solution. This is especially the case in the aforementioned largescale distributed environments. Building and maintaining robust hierarchical topologies in a distributed manner is the challenge this paper addresses. Local intelligent software agents play a central role in acquiring a global hierarchical topology using their ability to cooperate, adapt [4] and reconfigure. Local agent behavior can make the topology self-organizing. This paper focuses on tree topologies as an instance of hierarchical structures. In trees, local failures have a global impact in the topology as the removal of a node disconnects the branches underneath from the main body of the tree. Creating self-organized tree topologies that are resilient to failures requires: (i) to pro-actively organize (sort) agents over the tree in such a way that a potential failure has minimum impact on the tree structure, (ii) to re-actively adapt to changes in the environment by reconnecting, in case of failures, or rewiring connections to improve the robustness of the topology. These are the two main concepts of AETOS, the Adaptive Epidemic Tree Overlay Service. AETOS is the mechanism that this paper proposes to build and maintain robust tree topologies in distributed environments. It is based on a 3layer architecture as the core of self-organization. These three layers are facilitated and managed by a local software agent, the AETOS agent. Applications can be build on top of this agent-based dynamic tree overlay. The interaction between the AETOS agent and the application is managed through another local agent, the AETOS proxy. This agent provides the tree overlay on-demand to the application, and is responsible for bootstrapping and terminating the selforganization process. The contribution of this work is three-fold. This paper proposes the following: 1. The myopic competitive agent model, a highly reconfigurable self-organization model. It enables the use of dynamic proximity criteria. This model is inspired by
doi:10.1109/icais.2009.21 fatcat:4jaiapcdlrbcvgujze2nyiwcdy