Efficient and scalable exploitation of high-performance and parallel computing resources in massively parallel architectures or multicore systems [1] has been the origin of plenty of investigations in the related scientific communities. A genetic algorithm (GA) is a heuristic way to find a solution for an optimization or search problem [2]. A GA with its intrinsic iterative evolutionary processes potentially can be defined as a good candidate for solving problems with a concurrently solvable nature.
In the paper, the authors, by exploiting the parallel genetic algorithm (PGA) with its scalability, try to devise a solution for the generalized assignment problem (GAP). GAP can be considered as a model for capacity-constrained problems in a wide variety of domains. In this solution, the populations of the entities are divided into subpopulations, which are called demes. For the independent evolving demes, where their elements migrate into those nearby to constitute new generations, the paper expresses a scalable parallel search method to find the most fitted ones.
Asynchronous migration and buffer overflow resolution are the key lemmas to achieve scalability in the proposed method and algorithm. The discussion continues with the implementation mechanism of the offered scalable PGA. It declares the nodal interaction methods and expresses their related operations. Finally, great experimental results for the performance evolution of the proposed algorithm are provided.
Novel ideas concerned with demes evolution management in PGA are provided. The paper deploys policies to establish scalability, and with clear experimental results demonstrates their operational effectiveness and significance in related scientific domains.