Experimental Study of Optimization Algorithms for Resource Allocation in Cloud Computing

Abstract

Cloud computing is essential for executing scientific workflows, offering scalable resources to process large volumes of data. However, efficient resource allocation remains challenging due to fluctuating demand and inadequate planning, which can increase makespan and, consequently, energy consumption. In this context, Infrastructure as a Service (IaaS) is primarily used, enabling the rental of virtual machines (VMs) with different characteristics. This study identifies the factors that influence makespan improvement when executing workflows from diverse disciplines and with varying characteristics. The factors evaluated are workflow size, workflow structure, and the number of virtual machines required. The algorithms considered include four heuristics and one metaheuristic, specifically a genetic algorithm (GA). Among the factors, the most influential in cloud computing is the number of VMs: increasing the number of VMs reduces makespan up to a point, exhibiting the law of diminishing returns.