A Self-Optimization Method for System Service Dependability based on Autonomic Computing

Under the intrusion or abnormal attacks, how to supply system service dependability autonomously, without being degraded, is the essential requirement to network system service. Autonomic Computing can overcome the heterogeneity and complexity of computing system, has been regarded as a novel and effective approach to implementing autonomous systems to address system security issues. To cope with the problem of declining network service dependability caused by safety threats, we proposed an autonomic method for optimizing system service performance based on Q-learning from the perspective of autonomic computing. First, we get the operations by utilizing the nonlinear mapping relations of the feedforward neural network. Then, we obtain the executive action by perceiving the state parameter changes of the network system in the service performance. Thirdly, we calculate the environment-rewarded function value integrated the changes of the system service performance and the service availability. Finally, we use the selflearning characteristics and prediction ability of the Q-learning to make the system service to achieve optimal performance. Simulation results show that this method is effective for optimizing the overall dependability and service utility of a system.