Theses Doctoral

An Emergent Architecture for Scaling Decentralized Communication Systems (DCS)

Vicente, John Barbosa

With recent technological advancements now accelerating the mobile and wireless Internet solution space, a ubiquitous computing Internet is well within the research and industrial community's design reach - a decentralized system design, which is not solely driven by static physical models and sound engineering principals, but more dynamically, perhaps sub-optimally at initial deployment and socially-influenced in its evolution. To complement today's Internet system, this thesis proposes a Decentralized Communication System (DCS) architecture with the following characteristics: flat physical topologies with numerous compute oriented and communication intensive nodes in the network with many of these nodes operating in multiple functional roles; self-organizing virtual structures formed through alternative mobility scenarios and capable of serving ad hoc networking formations; emergent operations and control with limited dependency on centralized control and management administration. Today, decentralized systems are not commercially scalable or viable for broad adoption in the same way we have to come to rely on the Internet or telephony systems. The premise in this thesis is that DCS can reach high levels of resilience, usefulness, scale that the industry has come to experience with traditional centralized systems by exploiting the following properties: (i.) network density and topological diversity; (ii.) self-organization and emergent attributes; (iii.) cooperative and dynamic infrastructure; and (iv.) node role diversity. This thesis delivers key contributions towards advancing the current state of the art in decentralized systems. First, we present the vision and a conceptual framework for DCS. Second, the thesis demonstrates that such a framework and concept architecture is feasible by prototyping a DCS platform that exhibits the above properties or minimally, demonstrates that these properties are feasible through prototyped network services. Third, this work expands on an alternative approach to network clustering using hierarchical virtual clusters (HVC) to facilitate self-organizing network structures. With increasing network complexity, decentralized systems can generally lead to unreliable and irregular service quality, especially given unpredictable node mobility and traffic dynamics. The HVC framework is an architectural strategy to address organizational disorder associated with traditional decentralized systems. The proposed HVC architecture along with the associated promotional methodology organizes distributed control and management services by leveraging alternative organizational models (e.g., peer-to-peer (P2P), centralized or tiered) in hierarchical and virtual fashion. Through simulation and analytical modeling, we demonstrate HVC efficiencies in DCS structural scalability and resilience by comparing static and dynamic HVC node configurations against traditional physical configurations based on P2P, centralized or tiered structures. Next, an emergent management architecture for DCS exploiting HVC for self-organization, introduces emergence as an operational approach to scaling DCS services for state management and policy control. In this thesis, emergence scales in hierarchical fashion using virtual clustering to create multiple tiers of local and global separation for aggregation, distribution and network control. Emergence is an architectural objective, which HVC introduces into the proposed self-management design for scaling and stability purposes. Since HVC expands the clustering model hierarchically and virtually, a clusterhead (CH) node, positioned as a proxy for a specific cluster or grouped DCS nodes, can also operate in a micro-capacity as a peer member of an organized cluster in a higher tier. As the HVC promotional process continues through the hierarchy, each tier of the hierarchy exhibits emergent behavior. With HVC as the self-organizing structural framework, a multi-tiered, emergent architecture enables the decentralized management strategy to improve scaling objectives that traditionally challenge decentralized systems. The HVC organizational concept and the emergence properties align with and the view of the human brain's neocortex layering structure of sensory storage, prediction and intelligence. It is the position in this thesis, that for DCS to scale and maintain broad stability, network control and management must strive towards an emergent or natural approach. While today's models for network control and management have proven to lack scalability and responsiveness based on pure centralized models, it is unlikely that singular organizational models can withstand the operational complexities associated with DCS. In this work, we integrate emergence and learning-based methods in a cooperative computing manner towards realizing DCS self-management. However, unlike many existing work in these areas which break down with increased network complexity and dynamics, the proposed HVC framework is utilized to offset these issues through effective separation, aggregation and asynchronous processing of both distributed state and policy. Using modeling techniques, we demonstrate that such architecture is feasible and can improve the operational robustness of DCS. The modeling emphasis focuses on demonstrating the operational advantages of an HVC-based organizational strategy for emergent management services (i.e., reachability, availability or performance). By integrating the two approaches, the DCS architecture forms a scalable system to address the challenges associated with traditional decentralized systems. The hypothesis is that the emergent management system architecture will improve the operational scaling properties of DCS-based applications and services. Additionally, we demonstrate structural flexibility of HVC as an underlying service infrastructure to build and deploy DCS applications and layered services. The modeling results demonstrate that an HVC-based emergent management and control system operationally outperforms traditional structural organizational models. In summary, this thesis brings together the above contributions towards delivering a scalable, decentralized system for Internet mobile computing and communications.


  • thumnail for Vicente_columbia_0054D_10401.pdf Vicente_columbia_0054D_10401.pdf application/pdf 13.2 MB Download File

More About This Work

Academic Units
Electrical Engineering
Thesis Advisors
Campbell, Andrew T.
Ph.D., Columbia University
Published Here
May 22, 2013