Scalability demonstration of a large scale GPU-based

Large scale simulation is a challenging issue of the network research area. In particular, simulating one large space where a big number of nodes are in continuous interaction remains complex even if we consider distributed and parallel solutions. In this perspective; GPU appears as a promising hardware providing an important number of independent computing resources. Nevertheless its usage requires a new software design. In that context, Cunetsim is a distributed GPU-based framework which aims to combine the power of GPUs with the flexibility of distributed solution in order to increase the scalability while reducing the complexity. In this work we aim to demonstrate the efficiency and the scalability of that framework on one hand and its robustness in term of event handling on the other hand; therefore we propose a validation scenario including 1.5 millions nodes where we generate up to 10 billions events; we conduct the simulation using one workstation which includes three GPUs.