Section: Scientific Foundations
Models and methods for a self-management plane
Self organization and automation are fundamental requirements within the management plane in today's dynamic environments. It is necessary to automate the management processes and enable management frameworks to operate in time sensitive evolving networks and service environments. The automation of the organization of devices, software components, networks and services is investigated in many research projects and has already led to several solution proposals. While these proposals are successful at several layers, like IP auto-configuration or service discovery and binding facilities, they did not enhance the management plane at all. For example, while self-configuration of IP devices is commonplace, no solution exists that provides strong support to the management plane to configure itself (e.g. finding the manager to which an agent has to send traps or organizing the access control based on locality or any other context information). So, this area represents a major challenge in extending current management approaches so that they become self-organized.
Our approach is bottom-up and consists in identifying those parameters and framework elements (manager data, information model sharing, agent parameters, protocol settings, ...) that need dynamic configuration and self-organization (like the address of a trap sink). For these parameters and their instantiation in various management frameworks (SNMP, Netconf, WBEM, ...), we investigate and elaborate novel approaches enabling fully automated setup and operation in the management plane.
Design and evaluation of P2P-based management architectures
Over the last years, several models have emerged and gained wide acceptance in the networking and service world. Among them, the overlay networks together with the P2P paradigms appear to be very promising. Since they rely mainly on fully decentralized models, they offer excellent fault tolerance and have a real potential to achieve high scalability. Mainly deployed in the content delivery and the cooperation and distributed computation disciplines, they seem to offer all features required by a management framework that needs to operate in a dynamic world. This potential however needs an in depth investigation because these models have also many characteristics that are unusual in management (e.g. a fast and uncontrolled evolution of the topology or the existence of a distributed trust relationship framework rather than a standard centralized security framework).
Our approach envisions how a complete redesign of a management framework is done given the characteristics of the underlying P2P and overlay services. Among the topics of interest we study the concept of management information and operations routing within a management overlay as well as the distribution of management functions in a multi-manager/agent P2P environment. The functional areas targeted in our approach by the P2P model are network and service configuration and distributed monitoring. The models are to be evaluated against highly dynamic frameworks such as ad-hoc environments (network or application level) and mobile devices.
Integration of management information
Representation, specification and integration of management information models form a foundation for network and service management and remains an open research domain. The design and specification of new models is mainly driven by the appearance of new protocols, services and usage patterns. These need to be managed and exposed through well designed management information models. Integration activities are driven by the multiplication of various management approaches. To enable automated management, these approaches need to inter-operate which is not the case today.
The MADYNES approach to this problem of modeling and representation of management information aims at:
enabling application developers to establish their management interface in the same workspace, with the same notations and concepts as the ones used to develop their application,
fostering the use of standard models (at least the structure and semantics of well defined models),
designing a naming structure that allows the routing of management information in an overlay management plane, and
evaluating new approaches for management information integration especially based on management ontologies and semantic information models.
Modeling and benchmarking of management infrastructures and activities
The impact of a management approach on the efficiency of the managed service is highly dependent on three factors:
the distribution of the considered service and their associated management tasks,
the management patterns used (e.g. monitoring frequency, granularity of the management information considered),
the cost in terms of resources these considered functions have on the managed element (e.g. method call overhead, management memory footprint).
While the first factor was investigated in several research projects so far, none of the other two were investigated at all. The lack of such benchmarking data and models simply makes the objective evaluation of the operational costs of a management approach impossible. This may be acceptable in backbone networks where processing and communication resources can be tuned very easily (albeit sometimes at a non negligible cost). This is not true in constrained environments like devices constrained by battery or processing power as found in wireless networks for which the lack of management cost models is a serious concern.
MADYNES addresses this problem from multiple viewpoints: communication patterns, processing and memory resources consumption. Our goal is to provide management patterns combining several management technologies if needed so as to optimize the resources consumed by the management activity imposed by the operating environment.
Therefore, we establish abacuses for management frameworks and in parallel we collect data on current management practice. These data will form the core of the “Constraints-based management tuning activity” that we are working on and can be used for rigorous comparison among distribution and processing of management activities.