The purpose of this project is to show a methodology for the analysis of I/O workload of HPC clusters. We evaluate the I/O patterns of different I/O kernels in order to measure their influence on the workload of the HPC system. Our experiments were done in a Rock Cluster configured with a parallel file system Lustre. The experimental results showed that the applications require high computational resources, since the overload and possible errors in a HPC, installation and the execution of the several tests of the group of names, we clearly observe these problems in the architecture of our group at the time of access to the nodes, a cluster technique is used to identify overloads with interesting results.