EGAGP: An enhanced genetic algorithm for producing efficient graph partitions

Abstract

Graph partitioning is a well-known problem which has varied applications such as scientific computing, distributed computing, social network analysis, task scheduling in multi-processor systems, data mining, cloud computing and many other domains. In this paper, we propose EGAGP, an enhanced genetic algorithm for producing efficient graph partitions. Keeping real world applications in mind, our algorithm takes into account the capacity limitations of individual partitions to ensure balanced partitioning. This approach enables a large variety of applications for this algorithm, the most significant of which is in mobile cloud computing. Our algorithm can be used in efficient deployment of software components in cloud architecture as it is efficient and fast and it also ensures feasibility by only allowing partition sizes up to designated limits. We have achieved significant improvement over the previous state-of-the-art methods in terms of runtime and objective of graph partitioning cost. Our method is based on dividing the total execution time among primary and secondary populations and it resulted in an efficient algorithm. Several standard benchmark graph instances were used in our work to compare the performance of the algorithm. Our proposed method EGAGP is able to produce feasible and good quality results and outperforms the state-of-the-art methods in terms of time and quality of the solutions generated.