In the serial netCDF implementation, I/O can be a performance bottleneck because all the data must be scattered or gathered by a single task and that task alone communicates with the file system. In contrast, PnetCDF enables all tasks to communicate with the file system in parallel. When possible, PnetCDF offers further efficiency by combining multiple communication requests from one or more tasks into a smaller set of data transfers.