Optimizing the efficiency of manual parallelization

Today multicore technology is the standard processor technology for desktop and laptop computers and the software industry is facing the challenge of porting existing sequential code to parallel code. When parallelizing a sequential program most of the time a manual parallelization is necessary. One of the major obstacles for the software industry when porting existing applications are the high development costs. In this paper we propose a method which helps software developers to determine an efficient parallelization for a program in terms of speedup and effort associated with implementing the parallelization. First two optimization problems are established concerning the decision problem of selecting an efficient parallelization: speedup maximization and effort minimization. In order to define the concept of the parallelization a model for a manual parallelization will be developed. Based on this model a practical method is introduced how an efficient manual parallelization can be determined beforehand of implementation and also results of applying this method to the parallelization of a real-world code are presented. As a result we think that by adopting this method software developers can improve the efficiency of the manual parallelization leading to an improved cost efficiency of the parallelization process as a whole