Towards automation of IT systems repairs

Monitoring and repair are two sides of the on-the-fly maintenance of IT systems. Monitoring is well supported by automatic tools. In contrast, repairs involve much higher human intervention which negatively affects reliability and efficiency. The paper introduces a method of automating repairs of IT systems which can be integrated with any of the existing monitoring mechanisms. The method is described as a collection of models and algorithms. The monitored system is modelled as a multi-hierarchy of components, and predicates are used to detect problems and to trigger the corresponding repairs. Repair procedures are represented using flow diagrams. Graphical notations are used to specify the architecture of the runnable component of the proposed method and the corresponding repair API. Formal notation (the Z language) is used to provide for precision and unambiguity of the specifications. The formal specification facilitates reuse of the method in different target environments. For validation purposes, the method has been instantiated (using the PERL language) in an industrial helpdesk supporting the operation of several hundred IT systems. The validation data demonstrate the significant positive impact of the proposed method on the efficiency of the performed repairs.