Quantifying the Impact of Randomness in Simulation Based Studies

Abstract

Many scientific studies in Wireless Sensor Networks rely on simulations and correctness of these simulation results is heavily dependent on random numbers. In current situation, most of researchers generally use random numbers generated through common random number generating APIs of modern programming languages. This research activity describes the comparative analysis of existing random number generators and evaluates their impact of using different types of random numbers in a simulation based study of WSN. In this study, eight different types of random numbers generation algorithms are considered. These random numbers are first evaluated using standard random number testing procedures such as Run Test, Serial Test, and Chi square Test. After that the same random numbers in a Markov chain based probabilistic study of Wireless Sensor Networks are analyzed. Our empirical analysis reveals that there is a correlation between strength of random numbers and accuracy of simulation. Simulation provides correct results if the right Random Number Generator which is strong enough respect to its random properties is chosen. It is also shown that Random Number Generators with similar properties used in the simulation modeling produce similar results.