A Distributed Multithreaded Evolutionary Computing Frame Work using Differential Evolution Algorithm

Differential Evolution (DE), the evolution algorithm (EA) in the repository of evolutionary computing (EC), is known for its simplicity and adaptability to high performance computing paradigms. The distributed algorithmic frameworks of DE are the natural extension of DE, as DE is highly suitable for parallel and distributed computing. This paper proposes a study on implementing an enhanced distributed differential algorithm framework (named as mtdDE - multithreaded distributed DE) with multi-threaded islands in a distributed framework, to attain maximum level of data parallelism. The main objective of mtdDE is to reduce the computation time and to improve the solution quality for the targeted optimization problems. This paper presents this study in two phases. The Phase I implements the traditional distributed DE (dDE) framework and compares its performance with a state-of-the-art distributed DE, on a set of 8 benchmarking functions. In Phase II, the proposed mtdDE was implemented and its performance was compared with the dDE on a set of 4 benchmarking problems. The performance metrics considered were - the objective function values (ofv), the mean objective function values (mov), the execution time (et) and the average execution time (aet). The results acquired for the performance metrics of the dDE and mtdDE were studied, and the study revealed that the mtdDE outperforms the dDE in terms of both the solution accuracy and the computation time. The detailed discussion about the algorithmic structure, the experimental setup, the results obtained and the comparative study were presented in this paper.