Nowadays, cluster architecture is widely deployed, paving the path for grid computing. Sharing such high concentration of valuable resources requires special dedicated security mechanisms at various levels. Current solutions for autonomous systems are too intrusive to be deployed in a cluster. In order to develop new solutions adapted to the architecture, we consider a coupled approach for performing efficient anomalies and misuse detections in HPC environment. It relies on an instrumented node, and results in periodical sampling of applications, allowing the generation of specific security policies that can be used to perform further monitoring. This paper also deals with performance concerns and an approach is proposed in order to keep a tow overhead.