This bachelor thesis aims to explore different approaches of distributing a simulation on mobile devices organized in a peer-to-peer network structure. The starting point is a muscle visualization augmented reality iOS application, which tracks a person using the camera and superimposes a 3D arm model on the left arm of the person. The muscle contraction of the arm is visualized in two ways, either through color, or through the deformation of the model. The application is extended in order to distribute the arm simulation, using Apple's Multipeer Framework as the building block for the peer-to-peer network. Two measuring units which allow a comparison between mobile devices are introduced. The first evaluates a device's performance, while also taking into account the battery level. This unit is used to make a decision which device should perform inference using a neuronal network. This device then shares the results to other devices in the network. Using this approach, the computational load of the devices which do not use the neuronal network is reduced. The second evaluates the device's position in relation to the tracked person. It is used to decide either which device should run the neuronal network, or which device should perform the arm tracking of the person. With these two approaches, either the accuracy of the inference results or the tracking accuracy are improved. Distributing the simulation comes at the cost of an increased energy consumption, stemming from the necessary communication. The communication is performed over Wi-Fi, which is expensive in terms of energy usage.