Effciency evaluation of the input/output system on computer clusters

The increase of the complexity of scienti c applications that use high performance computing require more e cient Input/Output (I/O) systems. In order to e ciently use the I/O is necessary to know its performance capacity to determine if it ful lls applications I/O require- ments. This paper proposes the e cency evaluation of the I/O system on computer clusters. This evaluation is useful to study how di erent I/O system will a ect the application performance. This approach en- compasses the characterization of the computer cluster at three di erent levels: devices, I/O system and application. We select di erent system and we evaluate the impact on performance by considering both the ap- plication and the I/O architecture. During I/O con guration analysis we identify con gurable factors that impact the performance of I/O sys- tem. Furthermore, we extract information in order to determine the used percentage of I/O system by an application on a given computer cluster. Presentado en el XI Workshop Procesamiento Distribuido y Paralelo (WPDP) Red de Universidades con Carreras en Informática (RedUNCI)