Although compute nodes can perform computations on data simultaneously when ANSYS Fluent is running in parallel, when data is read from or written to a single, common file, the operations have to be sequential. The file has to be opened and read from or written to by processes that have access to the desired file system. It is often the case that the compute nodes are running on a dedicated parallel machine without disk space. This means that all of the data has to be read and/or written from the host process which always runs on a machine with access to a file system, since it reads and writes the case and data files. This implies that unlike the example in Message Passing Macros, where data is only passed to compute node-0 to be collated, data must now be passed from all the compute nodes to compute node-0, which then passes it on to the host node which writes it to the file. This process is known as “marshalling”.
The following sections describe the processes of reading and writing files in parallel in more detail: