If the Reader decompose option is available, then the reader for your chosen format is SOS aware. You would choose Reader decompose in the SOS options tab to task the data reader to read the data in partitions on each server. If this option is available, then you will usually choose it because the reader knows how to best decompose the data. This method requires that the reader be based on our 2.0 or later User-defined reader API, which is server-aware, and that it does its own determination of what data goes to each server. Some multifile and some multiblock formats for example, can easily allocate chunks of data by server. And they can do so based on the number of servers that are being used at the time. There is still the requirement that each server be informed of the same number of total parts. But the reader can return different portions of a given part based on which server is querying. And yes, a given part can be empty for some servers.

This type of partitioning reader has the advantage of being adaptable to the number of servers used. It does require the development of such a reader. Examples of that allow reader decompose are as follows: CGNS-XML, CTH, xdmf2, Exodus, Flow3d, Silo and VTK readers. In the data reader dialog (File → Open...) under the SOS options tab, are the SoS options for the selected data reader. One option is to pass wildcards to the server, for example if the data file refers to a set of files (file*.vtm).