Accelerating Trace Alignment Algorithm Applying Cuda

Abstract

Currently, the business processes that run businesses generate large volumes of traces. These tracesare stored in event logs for further analysis. The use of tools to extract useful knowledge from informationrecorded possible to know exactly happens in a company and the existence of anomalies executedprocess. The algorithm Trace Alignment allows to identify the most likely common behavior or executedprocess, the occurrence of deviations and the contexts in which one or more activities are carried out.Experiments show that the runtime depends on the number of traces desired alignment. The present articleintroduces techniques of parallel programming, CUDA, with the aim of reducing the execution time ofthe algorithm. The main features of the algorithm, as well as other parallel implementations are analyzedin order to unify the techniques that can achieve the best time. Parallel TA proposed algorithm decreasesabout 11 times, compared to the sequential implementation.