<![CDATA[Microservices architecture has become the dominant paradigm in modern application development; however, optimal load distribution and resilience against failures remain significant challenges. This study aims to empirically evaluate and compare the performance and fault tolerance of three load balancing strategies (Round Robin, Least Connection, and Weighted Round Robin) under varying load levels and controlled failure injection scenarios in a microservices architecture. The experiment was conducted using a Node.js and NGINX-based e-commerce system under three load levels with three replications per condition. The results show that under normal conditions, all three strategies deliver comparable performance (median difference < 5 ms). Significant differences emerge in tail latency (P99) at high load: Weighted Round Robin records the lowest (292.2 ± 13.7 ms), followed by Least Connection (313.1 ± 54.7 ms) and Round Robin (361.0 ± 67.4 ms). Under dual service failure (50% capacity loss), Round Robin's P99 spikes by 827%, while Least Connection and Weighted Round Robin remain stable (< 1%). Least Connection and Weighted Round Robin are recommended for production systems requiring high fault tolerance.]]>
Copyrights © 2026
