Failure Prediction for Critical Infrastructures

• Applicant: Professor Dr. Miroslaw Malek
• Subject Area: Security and Dependability, Operating-, Communication- and Distributed Systems
• Term: from 2006 to 2008
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 21501634

Failures are breakdowns where a system cannot deliver the intended functionality. The goal of failure prediction is to forecast these outages. While this problem has been addressed for non-distributed systems, this project focuses on the development of a model for failure prediction in distributed systems, especially for critical infrastructures such as the power grid. Based on our experience with failure prediction in complex software systems, we extend the scope in this proposal by investigating the prediction of failures in distributed systems. The prediction of failures is based on observations at multiple locations of the infrastructure. These observations are transferred to a failure predictor evaluating them according to what was observed and when and where it has been observed. The latter is the key difference to existing failure prediction techniques and it is one of the major indicators for dependencies between elements of the critical infrastructure. Learning about dependencies can help in preventing the spread of minor errors across the system that might lead to major failures. The need to analyze the when and where of observations in critical infrastructure systems leads to the second research goal of this project: The development of a data delivery middleware that is able to guarantee certain dependability properties including, e.g., maximum delay or availability. Additionally, the middleware will be composable in order to provide the flexibility to handle changes in thesystem configuration and to guarantee adequate data delivery in the case of failures.

• DFG Programme: Research Grants