The focus of typical architectural models is the description of large systems. Even though these systems are usually distributed, aspects of distributed systems are only addressed in a rudimentary way. While typical approaches pay attention to technical issues like deployment, data consistency problems are mostly ignored. This paper presents a modeling approach which integrates assumptions about data consistency into architectural models. The concepts of abstract locations and operations play a central role in this context. They allow transactions and snapshots to be implicitely defined by mapping high-level architectural models to low-level models. The decision about the application of transactional techniques and snapshot algorithms can then be based on architectural models. Furthermore, the approach can be integrated into programming languages and platforms. In this case, a runtime environment could automatically detect and process both transactions and snapshots.