Perancangan dan Implementasi Platform Web Load Testing untuk Evaluasi Ketahanan dan Performa Server

Abstract

The increasing reliance on web-based applications across various sectors has made server stability and responsiveness under concurrent user loads a critical quality concern. Conventional load testing tools such as Apache JMeter, while widely adopted as the industry standard, impose significant memory overhead due to Java Virtual Machine (JVM) dependencies and lack integrated real-time performance monitoring capabilities during test execution. This study presents the design and implementation of DwanLoadTest, an integrated web load testing platform that unifies scenario configuration, load execution engine, and a real-time analytics dashboard within a single system. The platform adopts a client-server architecture with a Next.js frontend and a Golang-based load engine leveraging non-blocking asynchronous goroutines and a circular buffer sliding window mechanism to ensure memory-stable operation. Performance evaluation was conducted by comparing metrics from DwanLoadTest against Apache JMeter 5.6.3 across five load scenarios ranging from 10 to 150 Requests Per Second (RPS) on three virtualized server targets with varying configurations: a Laravel-based server (PHP 8.0, PostgreSQL, 4 vCPU/8 GB RAM), a WordPress 7.0 server (MariaDB 10 MB, 4 vCPU/6 GB RAM), and a WordPress 6.8.2 server (MariaDB 108 MB, 6 vCPU/8 GB RAM), all hosted on Proxmox Virtual Environment 9.2. Experimental results consistently demonstrate that DwanLoadTest achieves superior resource efficiency across all three servers, consuming a maximum of 43.1 MB of RAM compared to JMeter which requires 435–782 MB an average ratio of 12.5x to 27.3x more efficient depending on the server configuration. The non-blocking goroutine architecture enables DwanLoadTest to distribute load traffic more precisely, recording a relative P95 latency improvement of up to 85.6% compared to JMeter at identical load levels. The platform successfully identified server saturation points across all three target configurations while maintaining a 0% error rate throughout all scenarios, demonstrating graceful degradation behavior in each environment. Endurance testing over five minutes confirmed the absence of memory leaks, with RAM growth of only 0.7 MB throughout the test duration. Cross-server analysis further revealed that application framework complexity and database size have a more dominant influence on server response time than hardware specifications alone. These findings establish DwanLoadTest as a lightweight, accurate, and operationally stable alternative for web server performance evaluation and capacity planning across diverse production infrastructure configurations.