Dominique Dudkowski

Probabilistic decentralized network management

This work proposes a probabilistic management paradigm for solving some major challenges of decentralized network management. Specifically, we show how to cope with 1) the overhead of redundant information gathering and processing, 2) the decentralized management in dynamic and unpredictable environments, and 3) the considerable effort required for decentralized coordination of management functions. To this end, we describe a framework for probabilistic decentralized management in the context of In-Network Management (INM). We demonstrate how this framework can be applied to a network of information, a novel clean-slate approach towards an information-centric future Internet. We show by means of a simulation study in the area of performance and fault management that we can significantly reduce the effort and resources dedicated to management, while we are able to achieve a sound level of accuracy of the overall network view.

Architectural principles and elements of in-network management

Recent endeavors in addressing the challenges of the current and future Internet pursue a clean slate design methodology. Simultaneously, it is argued that the Internet is unlikely to be changed in one fell swoop and that its next generation requires an evolutionary design approach. Recognizing both positions, we claim that cleanness and evolution are not mutually exclusive, but rather complementary and indispensable properties for sustainable management in the future Internet. In this paper we propose the in-network management (INM) paradigm, which adopts a clean slate design approach to the management of future communication networks that is brought about by evolutionary design principles. The proposed paradigm builds on embedded management capabilities to address the intrinsic nature, and hence, close relationship between the network and its management. At the same time, INM assists in the gradual adoption of embedded self-managing processes to progressively achieve adequate and practical degrees of INM. We demonstrate how INM can be exploited in current and future network management by its application to P2P networks.

Distributed Reallocation Scheme for Virtual Network Resources

Network virtualization is an emerging technology for cost-effective sharing of network resources. The key strategy in network virtualization is of slicing physical resources (links, CPU, memory, and storage) to create virtual networks that are assigned to different operators. One important challenge on network virtualization is the efficient use of the physical resources. To accomplish such efficient use the management of the physical resources should be transparent to the applications running within the virtual networks, and should be executed at runtime in order to deal with the variation on the load requests of different virtual networks. Traditional resource allocation schemes use offline, centralized, and global view strategies to manage the use of physical resources. In contrast to these strategies, we propose a runtime, distributed, local view approach to manage physical resources. In this paper we introduce a virtual network architecture and an associated self-organizing algorithm to reallocate virtual network resources along different physical nodes in order to equalize the bandwidth, and storage consumption on the physical nodes. We developed a virtual network model based on Omnet++ to simulate the designed self-organizing algorithm. An IPTV testbed scenario is presented and initial experiments, about the interruption time of the application inside the IPTV virtual network, are described.

A Framework for In-Network Management in Heterogeneous Future Communication Networks

Future communication networks will be composed of a diversity of highly heterogeneous network variants, ranging from energy constrained wireless sensor networks to large-scale wide area networks. The fact that the size and complexity of such networks will experience tremendous growth will eventually render existing traditional network management paradigms unfeasible. We propose the radically new paradigm of in-network management,which targets the embedding of self-management capabilities deep inside the network nodes. In this paper, we focus on our framework for in-network management,which allows management logic to be embedded and executed within network nodes. Based on a specific use-case of bio-inspired network management, we demonstrate how our framework can be exploited in a network failure scenario using quorum sensing and chemotaxis.

Event handling in clean-slate Future Internet management

Event handling is a management mechanism that provides means for the network to react on changes in the network conditions or performance. In the construction of a clean-slate management architecture, we consider this as a main building block. This paper proposes a fully distributed event distribution in a fully distributed environment: differently from existing works, no configuration is required in advance, and yet nodes have guarantee that events are delivered and that certain delivery objectives are respected. The contributions of this paper are: a generic system model for event handling and an analysis of event distribution mechanisms with respect to timeliness and traffic metrics. The paper describes and discusses in detail the results based on simulations and provides guidelines for management functions of the Future Internet.

A prototype for in-network management in NaaS-enabled networks

In-network management (INM) is a paradigm for distributed and embedded management for future networks. One of its main design goals is to be used in conjunction with Network-as-a-Service (NaaS)-enabled networks, which need to be managed efficiently, in a way that requires only little manual interaction, and across administrative network domains. In this paper, we present an elaborate an INM prototype that we have implemented to demonstrate INMs capabilities, based on our previously introduced INM architecture. Using a comprehensive network scenario, we discuss a number of real-time and algorithm measurements to demonstrate how INM can enable efficient management of NaaS-enabled networks.

Facilitating Adaptive Placement of Management and Control Functions in Converged ICT Systems

Mechanisms for the management and control (M&C) of largescale ICT systems, both established and innovative ones, generally follow a distinct approach on the dimensions from centralized to distributed and flat to hierarchical architectures. In this paper, we examine representative M&C frameworks and technologies and show that such a restrictive architectural choice is incompatible with system convergence, like computing/networking and fixed/mobile. To improve this situation, we propose a novel architectural approach that facilitates the adaptive placement of M&C functions by using different combinations of distribution and hierarchy patterns concurrently. Using a computing/networking systems scenario and a simulator prototype, we illustrate the potential of the proposed M&C approach in achieving more efficient overall M&C of converged ICT infrastructures.