An NSGA-III-Based Multi-objective Intelligent Autoscaler for Executing Engineering Applications in Cloud Infrastructures

Abstract

Parameter Sweep Experiments (PSEs) are commonplace to perform computer modelling and simulation at large in the context of industrial, engineering and scientific applications. PSEs require numerous computational resources since they involve the execution of many CPU-intensive tasks. Distributed computing environments such as Clouds might help to fulfill these demands, and consequently the need of Cloud autoscaling strategies for the efficient management of PSEs arise. The Multi-objective Intelligent Autoscaler (MIA) is proposed to address this problem, which is based on the Non-dominated Sorting Genetic Algorithm III (NSGA-III), while aiming to minimize makespan and cost. MIA is assessed utilizing the CloudSim simulator with three study cases coming from real-world PSEs and current characteristics of Amazon EC2. Experiments show that MIA significantly outperforms the only PSE autoscaler (MOEA autoscaler) previously reported in the literature, to solve different instances of the problem.