Simplicity, Reproducibility And Scalabilityfor Huge Wireless Sensor Network Simulations
Abstract
In this work we present two contributions for the wireless sensor network (WSN) literature.The first one is a general model to achieve reproducibility in kernel level of parallel simulators.Unfortunately, users must implement how their simulations repeat from scratch in WSN sim-ulators, but a parallel or distributed simulation imposes the concurrence principle, not trivialto be implemented by non-specialists. Tests using the simulator named JSensor proved thatthe model guarantees the most restrict level of reproducibility, even when simulations adoptdifferent number of threads or different machines in multiple runs. The second contribution isthe JSensor simulator, a parallel general purpose simulator for large scale WSN applicationsand high-level distributed algorithms. JSensor introduces more realistic simulation elements,such as the environment represented by customizable cells and application-events representingnatural phenomena, such as lightning, wind, sun, rain and more. The cells are placed in agrid that represents the environment with characteristics of the space defined by the users,such as temperature, pressure and air quality. Experimental evaluations show that JSensorhas good scalability in multi-core computer architectures, achieving a speedup of 7.45 in amachine with 16 cores with Hyper-Threading Technology, thus 50% of cores are virtual ones.JSensor also proved to be 21% faster than OMNeT++ while simulating a flooding model.
