An intelligent virtual machine allocation optimization model for energy-efficient and reliable cloud environment

The exponential growth of cloud computing has brought increased attention to energy efficiency in data centers. However, fluctuating resource demands and fixed virtual machine (VM) sizes lead to excessive energy consumption, inefficient resource utilization, and load imbalances. While dynamic VM consolidation mitigates these issues by reducing the number of active physical machines (PM), frequent consolidation can compromise system reliability, as VMs may be assigned to unreliable PMs. An effective resource management strategy is therefore essential for balancing energy efficiency and reliability in cloud data centers. This paper presents a novel resource prediction-based VM allocation approach that significantly reduces energy consumption while enhancing system reliability. The core innovation lies in optimizing a feed-forward neural network using the self-adaptive differential evolution algorithm, which integrates multi-dimensional learning and global exploration. Unlike traditional gradient descent algorithms, this method searches for the global best solution, offering more accurate and robust predictions of future resource usage. These proactive resource estimations enable fault-tolerant and reliable VM management, preventing system failures and improving overall performance. Evaluated with the Google cluster dataset, the proposed model outperforms existing methods, delivering remarkable reductions in power consumption (up to 44.81%) and the number of active PMs (up to 64.73%). Additionally, the system’s reliability improves by 65.25%, demonstrating the effectiveness of the approach in achieving energy-efficient and fault-tolerant cloud data center management.