Evaluasi Kinerja Load Balancing Berbasis Nginx pada Lingkungan Virtualisasi Menggunakan Proxmox

Abstract

The growing volume of web traffic on modern services, particularly e-commerce and video streaming platforms, calls for a reliable mechanism to distribute workload evenly across servers and prevent overload. Nginx, functioning as both a reverse proxy and load balancer, offers three widely used algorithms for this purpose: round-robin, least connections, and IP hash. This study evaluates the performance of these three algorithms within a Proxmox Virtual Environment (VE), an open-source, Debian-based virtualization platform supporting KVM and LXC. Testing was conducted across three scenarios, namely normal load, overload, and failover, measuring response time, throughput, CPU/memory utilization, and system recovery time. The results show that least connections consistently outperformed the other algorithms: an average response time of 141.0 ms (compared to 179.0 ms for round-robin and 167.7 ms for IP hash), a throughput of 107.7 requests/second (versus 93.3 and 98.0), the lowest resource usage at 63% CPU and 71% RAM, and the fastest recovery time of 3.6 seconds. An independent two-sample t-test at a 95% confidence level confirmed that the performance differences between least connections and the other two algorithms were statistically significant (p = 0.008 and p = 0.018), while the difference between round-robin and IP hash was not (p = 0.103). These findings suggest that a load-balancing mechanism that accounts for each backend server’s active connection load adapts more effectively than a static distribution approach, making least connections the recommended configuration for Nginx-based load balancing on the Proxmox virtualization platform.