Title: Management of Ad-Hoc Networks and Services
Author: Rémi Badonnel
e-mail: badonnel@loria.fr
Partner: INRIA
Supervisor: Radu State and André Schaff
Committee:
Year of start: 2003
Year of end: 2007
Funding institution: LORIA – INRIA Lorraine
This thesis addresses the issue of managing ad-hoc networks and their services. The management of these emerging networks cannot be efficiently achieved with existing management approaches for two major reasons. The ad-hoc network is spontaneously deployed from a set of mobile devices without requiring any fixed network infrastructure, so that the network topology may have an extremely high rate of change. Moreover, the network is self-maintained by devices that usually operate under many constraints such as limited battery power, restricted bandwidth and low system capacity.
We therefore propose a new management approach for ad-hoc networks capable to satisfy the multiple requirements in terms of integration, flexibility and low cost. This approach is organized into three main axes: (1) designing a management information model for ad-hoc networks, (2) reorganizing the management plane in a flexible manner based on a probabilistic scheme and (3) adapting the management operations to ad-hoc specifics based on analytical methods such as eigenvector centrality and k-means classification techniques. These management methods are evaluated through an extensive set of simulations and through the implementation prototype of a monitoring platform.
The first research issue I solved addressed the identification of the management information required for ad-hoc networks. The information model proposed by current management approaches is very insufficiently specified to support these dynamic networks. We therefore designed a complete information model capable to represent management data for the spatial, temporal and traffic aspects of ad-hoc networks. Thus, the accurate and comprehensive representation of parameters which drive the evolution of these networks, is made possible. We completely defined this information model in an homogeneous manner by extending the DMTF common information model (CIM). We also included an extension for managing the OLSR standardized ad-hoc routing protocol , similar to the existing extensions for routing protocols such as OSPF in traditional networks.
The second challenging issue consisted in designing the supporting management architecture. A pure management approach where all the nodes are managed at any time is too strict for ad-hoc networks. Instead of addressing the management of the whole network, we proposed at IEEE NOMS’06 a probabilistic scheme, with which the management task is limited to a specific subset of nodes in a light-weight and flexible way. With respect to such a selective management scheme, we detailed an algorithmic method to automatically assign manager roles to well-connected and network participating nodes. Probabilistic guarantees on the percentage of nodes to be managed were derived based on extensive set of simulations. We brought the benefits of our probabilistic scheme to the ANMP management architecture by designing a clustering extension at the application layer and by exploiting already available information provided by the underlying routing protocol.
The third research issue focused on adapting the management operations to ad-hoc network specifics. Our research work aimed at making them more robust to network mobility and less costly in resources. We defined at IEEE IM’05 a performance management method based on contrast filtering and graph dependency analysis. We also proposed at IFIP Networking’06 a fault management technique using information theory concepts. We experimented these management methods by simulation and implementation of a proof-of-concept prototype. These methods succeeded in evaluating the operational state of the ad-hoc network and to efficiently identify nodes that are particular with respect to presence, position and traffic in the network. This particularity can be positive (for instance, nodes that do important routing and constitute the equivalent of traffic highways in ad-hoc networks) or negative (for instance nodes that flood the network, without providing a minimum of service to it).
My long-term plans are to further investigate new methodologies and techniques to develop flexible, self-configurable and robust management systems capable to adapt to ever-changing environments.
Keywords: Network and Service Management, Ad-Hoc Networks, Information Model, Management Architecture, Performance Monitoring, Fault Detection
This thesis was defended by December 2006.The manuscript can be found in
http://tel.archives-ouvertes.fr/index.php?halsid=63eea5e4b0d3aa73dabd2a7e2c97896a&view_this_doc=tel-00134382&version=1