ParaView in Parallel
Architecture
The ParaView application can run on a whole cluster at the same time. It starts as many different processes, possibly on different machines. It has three main logical components, a client server, responsible for the user interface, a data server, to read and process data sets to create final geometric models, and a render server, which renders that final geometry.
The data server is a set of processes that communicate using MPI. On Stampede2 and Frontera, the data server runs on the cluster. It loads and executes the modules you select for your visualization. Each MPI task is told which partition, or piece of the data, it should load.
The render server, too, can run in parallel under MPI if it is configured to do so. There is a restriction that the render server cannot use more processes than the data server. Each render server receives geometry from data servers in order to render a portion of the screen. Spatial locality of geometry on the data server can reduce network traffic for this step.