Applied Edge Computing Simulation : the French Nuclear Power Plant Use case Cloud computing is the access to computing services (software, platforms, infrastructure, storage) trough internet from a provider. Today, cloud computing is the main strategy for data orchestration : data is collected trough sensors (smartphone, machine,…), send to a cloud that process the data and send back new data to the initial (or other) devices. This strategy is quite efficient if the application requires low data volume exchange and high latency . In the upcoming years, cloud computing won’t be efficient enough to manage applications like autonomous vehicles, smart grids/factory, mass VR, that generate more than 1 Go of data per second with extreme latency needed. For instance, a self-driving car must process 1 Go of data per second with a latency of 1/100s. The devised strategy to solve this problem is Edge Computing (EC). To put it simply, EC adds computing nodes between the devices and the cloud allowing efficient data orchestration with low latency . In the other hand, research on this subject are very challenging. This is due to three main factors:  the use of commercial infrastructures, such as Amazon, Azure or Google Cloud, or private experimental testbeds, affects the reproducibility of experiments  some measurements might not be available on commercial infrastructures, due to privacy issues  deployment and management of a real-world cloud infrastructure to perform experiments is either costly or time demanding. To solve this problem, simulators are being developed since a few years, allowing to compare and optimize cloud and edge architecture scenarios. In this conference, we provide a use case comparison of cloud and edge computing architecture for the overall French nuclear powerplants (26 powerplants, 56 reactors, 2000 sensors/reactors). 3 different strategies are simulated :  Tier 1 : Full cloud computing  Tier 2 : Edge computing with partial data processed on-site  Tier 2 with E0 : edge computing with distributed offload These strategies are tested on 6 different power plants data structure 1node/ n reactors (n=1,2,4), 1node/powerplant and 2 geographical cluster strategies. The first results show overall limitations in latency for the cloud architecture and possible optimizations for the nuclear power plant data orchestration.