Improved methods and measures for computing dynamic program slices in stochastic simulations

Stochastic simulations frequently exhibit behaviors that are difficult to recreate and analyze, owing largely to the stochastics themselves, and consequent program dependency chains that can defy human reasoning capabilities. We present a novel approach called Markov Chain Execution Traces (MCETs) for efficiently representing sampled stochastic simulation execution traces and ultimately driving semiautomated analysis methods that require accurate, efficiently generated candidate execution traces. The MCET approach is evaluated, using new and established measures, against both additional novel and existing approaches for computing dynamic program slices in stochastic simulations. MCET's superior performance is established. Finally, a description of how users can apply MCETs to their own stochastic simulations and a discussion of the new analyses MCETs can enable are presented.