Cloud computing opens up a variety of options, such as improving availability or fault tolerance of applications. However, it might also entail non-negligible overhead. There are different approaches for identifying and reducing overhead (e.g., by using a monitoring software). Most of those solutions are reactive and do not offer a way to do accurate predictions about scalability of applications running on cloud platforms. This thesis deals with the issue and considers the question of how to predict certain performance characteristics of an application without running it on the target platform. Generally accepted models such as Amdahl-s and Gustafson-s Law were examined for their applicability to cloud computing. Based on them, a model, which mathematically describes the scalability of applications under consideration of cloud specific properties, was developed. The model was evaluated on two applications with different load profiles. For this purpose, a lightweight profiler has been implemented to gather runtime information of the distributed applications. The collected data were filtered, clustered by thread, and aggregated by method and class level. Then they were interpolated and compared with the performance predictions. The deviations were analyzed and causes like garbage collection were discussed. The work shows that, for example, minor changes in the application can have a significant impact on the performance characteristics of the application. Models such as the one developed in the context of this thesis provide valuable information (e.g., scalability and technical limits).