Hybrid CPU-GPU distributed framework for large scale mobile networks simulation

Most of the existing packet-level simulation tools are designed to perform experiments modeling a small to medium scale networks. The main reason of this limitation is the amount of available computation power and memory in quasi mono-process simulation environment. To enable efficient packet-level simulation for large scale scenario, we introduce a new CPU-GPU co-simulation framework where synchronization and experiment design are performed on CPU and node's logical processes are executed in parallel on GPU according to the master/worker model. The framework is developed using Compute-Unified Device Architecture (CUDA) and denoted as Cunetsim, CUDA network simulator. In addition,

we propose a CPU-legacy version which is optimized for multi-core architecture.

In this work, we present cunetsim design models, its hardware/software architecture and supporting features. We evaluate the performance of cunetsim (both versions) compared to sinalgo and NS3 using existing benchmark scenario [25]. Evaluation results show that Cunetsim running time remains stable and that it achieves significantly lower computation time than CPU-based simulators for both static and mobile networks with no degradation in the accuracy of the results. To provide insights into the hardware configuration impact, we realize an experimental study of the simulation performance and correctness which demonstrates the importance of the cores' number for both versions. This detailed study allows us to suggest the most relevant configuration according to the simulated scenario and network.