A Lightweight Island Model for the Genetic Algorithm over GPGU

This paper presents a parallel approach of the genetic algorithm (GA) over Graphical Processing Unit (GPU) to solve Traveling Salesman Problem (TSP). Since earlier studies did not focus on implementing island model in persistent way, this introduces an approach, named Lightweight Island Model (LIM), that aims implement concept threads algorithm. For that, we present implementation details convert traditional model, which is separated into multiple kernels, computing paradigm based kernel. Many synchronization techniques, including cooperative groups and implicit synchronization, are discussed reduce CPU-GPU interaction existed model. A new parallelization strategy presented for distributing work among live during selection crossover steps. The GPU configurations lead best possible performance also determined. introduced will be compared, terms speedup solution quality, with (TIM) as well related works concentrated suggesting lighter version master-slave switching kernels (SAK) scheduled light kernel (SLK) approaches. results show can increase speed-up 27x serial CPU, 4.5x up 1.5–2x SAK SLK