An increasing number of applications operate in heterogeneous computing environments, often with mobile components. Methodologies that help developers assess the ability of such applications to meet their performance requirements throughout the software life-cycle are needed. In particular, early in the design phases, analysis techniques are critical for ensuring the future system's behavior, evaluating and comparing design alternatives. A performability evaluation is the most appropriate means to assess the expected system's ability to perform, including the effects of component failures and repairs. This paper focuses on model-based analysis of performability of mobile software systems. We propose a general methodology that starts from design artifacts expressed in a UML-based notation. Inferred performability models are based on the Stochastic Activity Networks notation. The viability of the proposed approach is demonstrated through its application in a case study.