Understanding autonomic network management: A look into the past, a solution for the future
Abstract
The evolution of mobile network technologies and their vertical integration, heterogeneity of applications, and the advent of sophisticated end-user devices have continuously been expanding the complexity of network management tasks. In addition, there is a significant urge for the dynamic reconfiguration of networks to meet operators costs and to achieve their performance objectives. These facts substantiate the idea of pushing the classical human dependent network management approaches out of the equation to a great extent. The vast scope of network management makes it difficult to have a common understanding and definition, which is often noticeable in different research articles. The situation is further worsened by the network evolution timeline that traverses several technological shifts, such as the time when computer networks and mobile networks were far apart, to the time of fully IP-based and converged networks. Hence, one of the main aims of this paper is to provide a study of the network management evolution in general and in particular the concepts of autonomic network management, so that researchers may be equipped to understand the involved concepts. To achieve the aforementioned objective, the authors carried out an elaborate analysis of the different network management approaches, mapped them to a timeline, and discussed their features. This analysis sets the stage for an extensive discussion of the enabling concepts of autonomic network management, followed by a survey of research projects targeting the advancement of the autonomic networking vision. Having identified incomplete realizations of autonomic network management due to simplifying assumptions, this paper focused on the relevant aspects of architectural construction with the presentation of the core challenges to be addressed so as to realize a fully autonomic network management framework. These challenges led us to reconstruct the design goals that the contributions of this work were built upon. The first proposal of this paper is to deploy intelligent software agents on different hierarchical layers of the proposed mobile network architecture. The agents implement different stages of cognitive control loops and contribute to learning algorithms for various management tasks. CoDIPAS-RL learning framework is used for layer specific learning decisions. To advance the autonomic network management, the authors also propose a novel idea of self-learning that enables the meta-learning vision. This paper concludes with a discussion on the implementation of our autonomic network management framework and with a use case that shows the performance of the proposed approach.