ML-WDLB: A Machine Learning–Based Dynamic Load Balancing Algorithm for Efficient Resource Allocation in Cloud Computing

Abstract

Cloud computing allows flexible and scalable computing resources. However, it is still a challenging problem how to efficiently distribute incoming workloads onto Virtual Machines (VMs). If the tasks are not well distributed, some VMs can be heavily loaded and others can be idle or lightly loaded. This can increase response time and processing time and can also reduce the overall utilization and performance of the cloud system. This paper presents a Machine Learning–Based Weighted Dynamic Load Balancing (ML-WDLB) approach for making VM selection more responsive to changing workload conditions. The approach uses both historical and current VM information, including CPU utilization, memory utilization, processing time, response time, and bandwidth. A Random Forest model is used to estimate the suitability of a VM for an incoming cloudlet or task. This prediction is combined with a weighted VM score that reflects the current condition of each VM. The VM with the highest final score, provided that it is below the specified load threshold, is selected for task execution. The performance of the approach can be assessed using response time, processing time, makespan, throughput, resource utilization, and load imbalance. The main aim of ML-WDLB is to provide a practical balance between prediction and current system information so that workload can be distributed more effectively as cloud conditions change.