There are many ways to define the term scalability. For most users of the Mechanical APDL program, scalability generally compares the total solution time of a simulation using one core with the total solution time of a simulation using some number of cores greater than one. In this sense, perfect scalability would mean a drop in solution time that directly corresponds to the increase in the number of processing cores; for example, a 4X decrease in solution time when moving from one to four cores. However, such ideal speedup is rarely seen in most software applications, including the Mechanical APDL program.

There are a variety of reasons for this lack of perfect scalability. Some of the reasons are software- or algorithmic-related, while others are due to hardware limitations. For example, while a DMP solution can run on as many as 1024 cores, there is some point for every analysis at which parallel efficiency begins to drop off considerably. In this chapter, we will investigate the main reasons for this loss in efficiency.