An execution flow for dynamic concurrent systems: simulation of WSN on a Smalltalk/CUDA environment.
Résumé
The progress in wireless sensor networks is the driving force behind the idea of achieving ambient intelligence through pervasive computing and robotics. Sensor networks are fault-prone and must have high energy efficiency. Furthermore, the applications are event-driven distributed algorithms requir- ing high performance in bursts. Hence, high-level synchronous simulation is a necessary step for programming and analyzing such systems in terms of scalability, reliability, and other design objectives. In this paper, we introduce a simulation framework for dynamic networks in Smalltalk. Its originality lies in using massive parallelism of Graphic Processing Units (GPU) through dynamic linked libraries (DLL), for accelerating the synchronous simulation. Another important contribution is the capability of the framework to handle mobile sensors interacting with a static sensor network. The results of the paper show the feasibility of the GPU acceleration approach by showing the fast simulation time of a case study defining mobile interaction with dynamic network. A result of this work is a tight coupling of a Smalltalk VM with a set of process networks that can bring parallel execution opportunities to the virtual machine.