Abstract-Peer-to-Peer Realm (P2PRealm) is an efficient peer-topeer network simulator for studying algorithms based on neural networks. In contrast to many simulators, which emphasize on detailed network simulation, the speed of simulation in P2PRealm is essential, because neural networks require a time consuming training phase. Efficiency has been obtained by optimizing training loops inside the simulator, using Java Native Interface (JNI) as well as distributing the simulator to hundreds of workstations using the P2PDisCo platform. In this paper we describe the architecture of P2PRealm and its input/output interfaces. Also, we present the mechanisms used for internally optimizing the implementation and the configuration used for distribution. Finally, we present the use of P2PRealm with the P2PStudio network visualization tool.