Агентный подход к реализации модели поведения пользователей Grid-систем

Abstract

We propose an agent approach to the implementation of model of users behavior in Grid systems. Proposed model is based on the analysis of statistical data about jobs that were completed by the user in Grid system. In order to distinguish normal and abnormal user’s behavior we use feed-forwards neural networks. We verified the proposed approach on data that were collected by GridICE monitoring system in GILDA-EGEE. The results of experiments showed that our model was able to distinguish abnormal user’s behavior in 90%.