Online services have become an indispensable part of our lives. The Internet of Things (connecting electronic devices to the Internet) has added another possible application in recent years. The resulting increase or fluctuation in the number of users means that these services must remain available and accessible. Disruptions and outages have serious consequences. If, for example, the scaling of a service does not function properly, the provider may incur considerable costs. Cloud engineers work with so-called scaling policies to enable elastic systems that automatically scale resources above a certain threshold of a metric. To avoid mistakes, architecture-based simulations like Palladio’s can help. Palladio’s effectiveness has been demonstrated in various scenarios such as software as a service. However, whether Palladio is also effective in the context of scaling policy has not been sufficiently investigated so far. The goal of this work is to investigate the feasibility of scaling policy simulation using Palladio. Subjects of investigation are the accuracy of the simulations and whether the Palladio model helps with the comprehensibility of scaling policies. To determine the accuracy, measured values are recorded during an experiment and later compared to the Palladio simulation results. A Kubernetes cloud system from the MoSaIC project is used as a reference system. The use case of the project is container ships sending data, with the number of ships increasing due to a new customer. To generate the load, the load testing software Gatling is used. The experiment is divided into two phases, a scaling experiment and an elasticity experiment. The former is used to quickly rank the MoSaIC Kubernetes system, which is a prerequisite for the design of the elasticity experiment. The design provides two load scenarios (low and medium). With these scenarios, two different scaling policy configurations, which differ in terms of the threshold value, are put under load in the experiment. These scenarios, the system, and the scaling policies were modeled with Palladio. During the modeling process, we found that various factors made it difficult to model the experiment scenario and run the simulation. The corresponding deficiencies and knockout criteria and possible workarounds to circumvent the problems were documented. The scaling policies could not be simulated to the full extent. Therefore it was not possible to simulate them. However, we were able to show the potential of Palladio and that the Palladio model we used allows the tracking of how self-adaptations were performed. That theoretically improves the understandability of the scaling policies. Future work can build on our findings and find out via further experimentation whether the documented deficiencies can be fixed or circumvented. In addition, an experiment should be conducted to investigate whether the improved understandability of the scaling policies through Palladio can also be proven.