Model-based decision support of task allocation in global software development

Global Software Development (GSD) has become a common practice in industrial software engineering: Often, teams from distributed sites all over the world cooperate in order to produce a software product. While the rise of GSD is driven by the expected benefits it promises to deliver, such as access to a global talent pool or low labor cost rates in distant regions, GSD also imposes many new problems that imply severe project risks. These problems are mainly caused by barriers in communication, coordination, and control, as well as by differences in expertise and background between sites. One central question in GSD project planning is how to allocate work across distributed sites, as this has significant impacts on the severity of GSD-specfic challenges and risks. In practice, however, such assignment is often not done systematically: Usually, there exists neither a method for work allocation that considers the multiple relevant influencing factors nor any systematic knowledge regarding the effects of different assignment possibilities. Experiences from previous GSD projects are not systematically reused in the task assignment decision. Therefore, the impact of work allocation on GSD-specific project risks is often not considered. As a result, risks are overlooked in many GSD projects, causing reduced productivity and lower project success rates. This thesis presents a method for the model-based evaluation and selection of work allocation alternatives. The main elements of the method are three models: one for suggesting work allocation alternatives, one for evaluating them with respect to effort and cost, and one for assessing project risks. These three models are connected via a common causal model that is able to store organization-specific experiences in a structured way. The method provides (a) a formal description of the different models and transformation rules between them, (b) processes for organization-specific instantiation of the models and for applying them during project planning, and (c) an implementation of the models in a prototype tool. Further contributions include an empirical interview study of the state of the practice in GSD and an overview of the state of the art in task allocation. The approach was evaluated in two studies with experts from industry and by using data from historical industrial projects. In each study, organization-specific instances of the models were developed and evaluated. One result of the evaluation is that the approach is able to accurately identify the impact of work allocation on GSD-specific project risks: More than 67% of the identified risks could actually be observed in the projects. In addition, allocation decisions suggested by the approach were seen as reasonable by experienced practitioners.