Parallel I/O for the CGNS system

Linux PC Clusters are a cost-effective platform for parallel computational fluid dynamics (CFD) applications. However, obtaining high perfomance on these clusters is a non-trivial problem, requiring tuning and design modifications to the code (and, in some cases, the hardware itself). In this paper we present our investigations in optimizing I/O performance using the portable CGNS file format. This format is becoming a standard interface for writing grid and solution information in a portable way. However, the original interface does not provide a mechanism for parallel data storage and access. In this work a parallel interface for writing and reading CGNS datasets based on the parallel HDF5 interface is described. Our tests indicate programming convenience and I/O performance improvement with the parallel CGNS interface.