Virtual Machine (VM) consolidation and migration are crucial for achieving resource efficiency and reducing operational costs in large-scale cloud data centers. Although researchers have proposed a variety of VM selection and placement strategies (including Particle Swarm Optimization, or PSO), many existing works focus on CPU-based thresholds alone, offer limited complexity analysis, and lack extensive experiments for validation. In this paper, we propose an enhanced PSO-based framework that integrates both CPU and memory utilization thresholds in the SLA definition, employs a more selective VM migration metric, and includes a thorough time-complexity and convergence discussion. We conduct experiments across three different data center sizes (100, 150, and 200 PMs) with up to 1000 VMs. Our results show that the proposed method reduces energy consumption by up to 23% compared to the Modified Best Fit Decreasing (MBFD) heuristic, while also minimizing SLA violations and load imbalance. We further present a detailed parameter tuning and sensitivity analysis ( α ,β, w 1 , w 2 , w 3 ), discuss scalability, and compare our approach with recent fuzzy and deep reinforcement learning (DRL) solutions from 2022-2026. The outcome is a robust, flexible solution that can adapt to multi-resource constraints and evolving workload patterns.
Khanday et al. (Thu,) studied this question.