A Comprehensive Review of A Proactive Auto-scaling and Energy-Efficient VM Allocation Framework Using an Online Multi-Resource Capsule Shuffle Attention Network for Cloud Data Centres

Abstract

Cloud computing has become a fundamental technology for delivering scalable computing resources and services across distributed environments. Modern cloud data centres host thousands of virtual machines (VMs) that dynamically process large-scale workloads generated by web applications, data analytics platforms, and enterprise systems. However, the rapid growth of cloud services has significantly increased the complexity of resource management and energy consumption within data centres. Efficient resource allocation and dynamic auto-scaling mechanisms are therefore essential for maintaining service quality while minimizing operational costs and power consumption. Virtual machine allocation and auto-scaling play a crucial role in cloud infrastructure management. Improper allocation of VMs can lead to resource underutilization, increased energy consumption, and violation of service level agreements (SLAs). Energy consumption in cloud data centres has become a critical issue due to the massive number of servers required to support modern applications. Studies indicate that optimizing VM placement and consolidation strategies can significantly reduce energy usage while maintaining system performance. To address these challenges, researchers have proposed various optimization and machine learning techniques for cloud resource management. Recently, proactive auto-scaling frameworks have been introduced to predict future workload demands and allocate resources accordingly. These frameworks rely on predictive models to forecast resource requirements and dynamically scale VMs before performance degradation occurs. For instance, neural-network-based models have been used to forecast multi-resource demands such as CPU, memory, and storage requirements simultaneously.