Significant problems facing all experimental and computationalsciences arise from growing data size and complexity. Common to allthese problems is the need to perform efficient data I/O on diversecomputer architectures. In our scientific application, the largestparallel particle simulations generate vast quantities ofsix-dimensional data. Such a simulation run produces data for anaggregate data size up to several TB per run. Motived by the need toaddress data I/O and access challenges, we have implemented H5Part, anopen source data I/O API that simplifies the use of the Hierarchical DataFormat v5 library (HDF5). HDF5 is an industry standard for highperformance, cross-platform data storage and retrieval that runs onall contemporary architectures from large parallel supercomputers tolaptops. H5Part, which is oriented to the needs of the particlephysics and cosmology communities, provides support for parallelstorage and retrieval of particles, structured and in the future unstructuredmeshes. In this paper, we describe recent work focusing on I/O supportfor particles and structured meshes and provide data showing performance on modernsupercomputer architectures like the IBM POWER 5.