Configuration of Mail Delivery System Based on Reinforcement Learning

To solve the internal security problem such as human error and bad faith, the automation of computer system management is significant. For this purpose, we focus attention in the automation of Mail Delivery Service. Today, requirement for reliable mail delivery system becomes larger and larger. However, existing systems are too strict about their configuration. Hence, we propose the method based on Reinforcement Learning (RL) to achieve proper MX record ordering. A measure on validity of the design of system, such as network topology number of servers and so on, is also obtained as side benefit. In order to verify the usability of the presented method, we did on a small model of mail delivery system. As a result, we show that RL is available for determination of the proper MX record ordering. Additionally, we suggest that it is also available for comparing validity of setting of MTA and the network design.