This paper presents a methodology for selection of optimalsoftware design parameters using the experimental design. When anorganization is at the point of taking up a new project with anobjective of improving the software quality, Taguchi method is appliedfor the software design process with an objective that not more than oneerror is found per software module. The strategy in robust design is toconduct off-line experiments using orthogonal arrays (OA) and tooptimize the design by maximizing performance measures with respect todesign parameters. Towards this a cause and effect diagram for designerrors was drawn with opinions from customer, production, qualitypersonnel and engineers. This diagram gave three most likely parametersas candidate for software design error, they are coupling, number ofrequirements per module and McCabe's cyclomatic complexity. It wasplanned to consider coupling parameter at two levels as low coupling atlevel 1 and high coupling at level 2, in case of number of requirementsper module parameter three levels were considered they are onerequirement per module at level 1, two requirements per module at level2 and greater than two requirements per module as the level 3. In caseof McCabe's complexity value<5 is set at level 1, value 5 to 10 isset at level 2 and value >10 is set at level 3. The possible numberof factorial experiment required be conducted for levels selected forthe three parameters is 18. The appropriate orthogonal array based onthe guidelines of Taguchi is L9 that is nine experiments needto be conducted to find the optimal software design parameters
展开▼
