A Comparative Study of Dynamic Load Balancing Algorithms for Microservices in Heterogeneous Multi-Cloud Environments

Abstract

Microservices-based application architectures in cloud environments require load balancing mechanisms that can adapt to differences in server capacity and workload fluctuations. This study aims to evaluate the performance of dynamic load balancing algorithms—Least Connection, Weighted Least Connection, and Least Response Time—in heterogeneous server environments using Amazon Web Services (AWS) and Google Cloud Platform (GCP). The evaluation was conducted through staged testing by observing application performance based on p95 latency, throughput, error rate, and load distribution patterns. The results indicate that no single algorithm consistently outperforms the others across all scenarios and platforms. Weighted Least Connection tends to produce a more proportional load distribution according to server capacity, while Least Connection and Least Response Time are more influenced by the number of active connections and initial response time. Overall, both AWS and GCP are able to maintain application performance stability across all load levels. These findings confirm that the effectiveness of dynamic load balancing algorithms in heterogeneous cloud environments is influenced by workload characteristics and server capacity, indicating that algorithm selection should be tailored to the specific system objectives.