Data Model

 The intuitive application and system models must be brought into an integrated canonical integer form to make use of math programming. We assume that mathematical programming is efficient enough for loadtime execution if the state space is sufficiently constrained. We experimentally verify this point in the evaluation section. Model building is more art than science, but structured approaches do exist. We follow the classic approach of structured synthesis [Sch81]. Here, an optimization problem is formulated as built from four datasets. The first, variables, implement choices, such as which channel to select. The second, parameters, numerically encode properties, such as a channel's capacity. Constraints limit the solution space; they are expressed in linear (for a linear program) relationships of variables and parameters. Finally, an objective also expresses a relationship of variables and parameters, but instead of an (in)equality, it is a function to be minimized or maximized (i.e., optimized). Goal is to find the set of variable values that achieves this. This section first introduces each dataset in detail and then summarizes the integrated model. We do not explain the concept of mathematical programming here; the approach is introduced independently from application tailoring (with simpler examples) in Appendix BAutomation Algorithms.