next up previous

4.2 Hierarchical Data Mapping

Some applications involve many computations with data elements which are near to each other in the data set; in such a case hierarchical data mapping is a good choice, since it maps data points which are adjacent in the program's data file to the same processor. This method of data mapping is implemented by dividing the data set into as many sections as there are processors and assigning one section to each processor. On pulsar, which has 4096 processors, the data set would be divided into 4096 sections.

Consider the data set below, where represents the data element:

Suppose the machine in use has 6 processors. With the hierarchical strategy we could assign data to processors as follows: