Re-engineering Sequential Software to Introduce Parallelization

Abstract: In the quest for additional computational power to provide higher software performance, industry have shifted to multi-core processing units. At the same time, many existing applications still contain sequential software; in these cases, multi-core processors would not deeply improve performances and in general would be under-utilized since software running on top of them are not conceived to exploit parallelization. In this thesis we aim at providing a way to increase the performance of existing sequential software through parallelization and at the same time minimizing the cost of the parallelization effort. The contribution of this thesis is a generic parallelization method for introducing parallelization into sequential software using multi-core CPUs and GPUs. As a proof-of-concept we ran an experiment in industrial settings by applying the proposed parallelization method to an existing industrial system running sequential code. Additionally, we compare the method we propose to existing methods for parallelization.