Learning to Enhance Dependability of Information Systems

Automatic configuration of large and heterogeneous ICT systems and their dependability mechanisms is both desirable and daunting for the inherent complexity of these systems.Configurations are commonly designed based on personal expertise, best practice, empirical evidence, without any automatic process and formal validation mechanism. This approach leads to frequent and reiterate errors with severe impacts on system reliability and security.In this article we present a general methodology and a set of tools to generate and validate the configuration of complex ICT systems, with little human intervention. These abstract configurations are to be used to drive the work of semi-automatic system management frameworks and tools that directly interact with the live system.We start from a formal description of the system including its service and infrastructure parts, software and hardware components, and associated dependability features. Through automatic tools we generate a set of possible system configurations that satisfy a set of security and reliability constraints. These tools exploit automated learning techniques to self tuning and customize their behavior based on feedbacks coming from the live system via its management and monitoring infrastructure.