OPTIMIZING CLOUD RESOURCE ALLOCATION USING MULTI AGENT DEEP REINFORCEMENT LEARNING-BASED ADAPTIVE HHO ALGORITHM

Abstract

Cloud computing workloads are projected to grow by 23.1% annually, with over 80% of enterprises 
adopting multi-cloud strategies, creating a pressing need for optimal virtual machine (VM) resource 
allocation to ensure cost efficiency and performance. However, existing allocation strategies suffer 
from static optimization limitations and inefficient adaptation to dynamic workloads, leading to 
frequent resource underutilization and service delays. To address these issues, this work proposes a 
novel Multi-Agent Deep Reinforcement Learning-Based Adaptive Harris Hawks Optimization 
(MADRL-AHHO) algorithm for cloud resource allocation using the VM Resource Allocation dataset 
(VM-0 to VM-5 classes). Initially, the dataset is preprocessed through normalization and feature 
selection to reduce dimensionality and noise. Feature extraction is enhanced using a Convolutional 
Neural Network (CNN) integrated with Firefly Optimization (CNN-FFO), which is benchmarked for 
learning capacity and solution convergence. However, performance limitations in CNN-FFO under 
dynamic load conditions are overcome by integrating CNN with Adaptive Harris Hawks Optimization 
(CNN-AHHO), which dynamically adjusts its exploration and exploitation capabilities based on 
multi-agent reinforcement feedback. The agents interact with the environment to learn optimal VM 
allocation policies by maximizing resource utilization and minimizing SLA violations. Experimental 
results demonstrate that CNN-AHHO outperforms CNN-FFO and traditional machine learning 
methods in terms of allocation accuracy, convergence rate, and computational efficiency, thereby 
offering a robust and adaptive solution for cloud infrastructure management.