Providing PVM with a parallel file system for cluster grids

Cluster grid computing is nowadays considered an actual alternative to both grid computing and traditional parallel computing based on supercomputing systems and on clusters of workstations exploited as a unique, coherent, high performance computing resource. To this end, ePVM is an extension of the well known PVM programming system, whose main goal is to enable PVM applications to run on "cluster grids". However, for large-scale PVM applications to be able to efficiently run on such computational grids, ePVM needs to be provided with a parallel file system that enables the enormous volumes of data usually generated by such applications to be managed and distributed among multiple disks. This paper presents the optimized porting of the PIOUS parallel file system, called ePIOUS, under ePVM. The porting has been carried out taking into account the two-levels physical network topology characterizing the cluster grids normally built by ePVM. Furthermore, in order not to penalize the application performance, ePIOUS has been also provided with a file caching service that is able to speed up file accesses across clusters.