Estimation of parallel algorithms efficiency based on modified Petri nets

Parallel computers have a significantly higher limiting speed of processing compared to sequential ones. However, it is not always possible to realize the potential of parallelism in practice. It can be due to the difference in the types of parallelism between the algorithm and the computer, the inability to parallelize the algorithm onto given number of executive devices or due to non-optimal representation of the program in the computer code. The authors propose an approach to analyzing of the time characteristics of parallel algorithms. They also describe the particular case often missed in considering when the overhead for parallelization may exceed the effect of parallelization. The proposed method allows to assess the influence of indirect factors of estimating and accounting for the overhead costs for the execution of a parallel algorithm.