Toward an experimental method for algorithm simulation

This feature article surveys issues arising in the design, development, and execution of computational experiments to study algorithms. An algorithm is viewed here as an abstract model of an implemented program: experiments are performed to study the model, and new insights about the model can be applied to predict program performance. Issues related to choosing performance measures, planning experiments, developing software tools, running tests, and analyzing data are considered. Some hazards and difficulties of computational research that arise particularly in the context of algorithmic problems are also surveyed.