A parallel and distributed multi-population GA with asynchronous migrations: energy-time analysis for heterogeneous systems

Speed and energy efficiency are two concepts that currently should be taken into account when creating parallel code, especially in time-consuming applications such as GAs. Although the performance of single-computer systems continues to increase, it does not improve at the same rate as the computing requirements do. The depletion of Moore's law, the high frequencies that microprocessors already reach, or the difficulty in continuing to reduce lithography are some of the causes that make single-computer systems not suitable for many applications. In response, distributed systems emerge to overcome hardware limitations and meet the requirements of the applications. With this in mind, this paper provides an efficient multi-population GA with asynchronous migrations and parallelism at multiple levels by exploiting the capabilities of a heterogeneous four-node cluster. The procedure is evaluated from an energy-time point of view and compared to a synchronous version. The results show the importance of developing efficient methods to achieve good performance and demonstrate that energy-aware computing is the way to continue on the right track.