In this section we considered divisible tasks, i.e. parallel applications that can be arbitrarily divided and executed in parallel. The analysis usually included two steps: devising a scattering algorithm and solving a set of linear equations. The former covered the underlying hardware/software architecture. The latter included solving two types of equations: equations linking processing time and communication time of the sender and the receiver, and an equation expressing that all the load must be processed. This method has been applied successfully to analyze many different types of computer architectures.
We assumed a linear relation between the processing time and the volume of data. Yet, even distributed sorting has a nonlinear dependence between the processing time and the size of processed data. Such nonlinear dependencies can be included in our equations (e.g. (1), (5), (10), etc.) as a nonlinear processing time function of the amount of assigned load. However, in this case the equations would be more difficult to solve.
Let us also mention that the divisible task approach can be applied in production - transportation systems. In such systems the transportation system is an equivalent of the computer interconnection network, while production facilities represent processors.