Engineering Methods Research For Software Process Performance Model

Software performance model is the major method to predict and control software process in high maturity companies. It is also a difficulty in implementing CMMI for most of the companies. 5 key difficulties they often run into when they establish software performance model:(1)Misunderstanding the Software performance model,often take predictable model as performance model, and take determinative model as performance model.(2)Have some historical data, but without high quality, there are many mistakes in data. Wrong conclusion will be output if PPM is created based on the wrong data.(3)Do have historical data, but the x and y are not best representatives, or quantitative relationships cannot be established as the processes are not stable.(4)Correct method of establishing PPM is not used so that the PPM makes no sense in practice or PPM cannot be established.(5)The meaning, category and way of establishing PPM is well defined in this article in combining lots of practices of many companies.In this article, several examples are given to demonstrate how to establish PPM according the theory and practice. I also provide several ways to demonstrate how to identify Y and X, they are very practical.The feature, category and use of PPM are characterized, and deeply research of PPM's concept is analyzed in this article. Retrospection the brief history of PPM, comprehensive description of original PPM and method used in software processes are described in this article also. GQM and process attributes analysis method are mentioned here as many people often ignore the establishment of x and y. Engineering method, points for attention and PPM examples are described including line regressive analysis, ANOVA, MONTE CARLO simulation and Bayesian Network. Resolution of these issues are given for these three methods, from now on, the establishment of PPM is deployed from theory to practice.The 3 major achievements are:(1) The methods to identify dependent and independent variables in process performance model are mentioned;(2) Four methods are established to define process performance model in this article;(3) Ways to forecast and optimize the process performance models are given in this article;In this article, we firstly describe the features, category, and usage of the process performance models and then give the detailed information of them. Review the history of process performance model and discussed the methods to help the readers to create them by regression analysis, ANOVA, MONTE CARLO simulation and Bayesian Network. I also provide the advantages and disadvantages of the each method.The methods mentioned in this article are deployed in Changhong, CVIC software factory in Kunshan and Hoperun in Nanjing. They are proved very effective and efficiency is also proved in practice.