Modeling of Software Partition for Distributed Real-Time Applications

The issue of software partition deals with the process of mapping the given set of logical modules, which reflect the user's point of view, into a set of software tasks, which reflect the software implementor's point of view. It is shown in this paper that the software partitioning problem can be modeled as one that maximizes the efficiency in resource utilization while observing the constraints on CPU throughput, memory space available, maximally allowed task execution time, and the order of module execution. The CPU and memory constraints are implementation dependent. The maximum task execution time constraint is due to considerations on the response time performance. The constraint on module execution order is a logical one, and it is shown to have significant performance impact. It is proven that by employing the module precedence relation, which reflects the sequence of module execution, the order of module execution can be properly maintained during the software partitioning process. And thus the defined tasks can be guaranteed to be completely executable, once properly activated With completely executable tasks, the operating overhead cost and the response time delay can be minimized. The following four module precedence relations are explored: precede, succeed, parallel, and precede as well as succeed. The validity of the selected partitioning criterion of maximizing the resource utilization efficiency is also assessed through simulation experiments. The results of simulation show that performance of the selected criterion is insensitive to the application environment, as well as to the application requirements.