Benchmarking Traditional Web Servers and Container-Based Serverless Platforms for a Django E-Commerce System

Abstract

We benchmark a Django 5.2 e-commerce application (Python 3.11, SQLite3 as a controlled database constant) on four hosting configurations – Apache 2.4 on Ubuntu 22.04 via WSL2, Nginx 1.27.2 on Windows 11, Azure Container Apps (ACA), and Koyeb – under five load profiles (C = 10–10,000; n = 500–50,000 requests). Sixteen harmonized performance metrics were evaluated, covering mean and tail latency, throughput, error rate, resource utilization, energy-per-request proxy, security timing, and four cloud elasticity metrics. Under the local loopback topology, Apache achieved lower client-observed latency and higher throughput; Nginx exhibited zero errors and tighter tail spread across all load levels. In the cloud topology, Koyeb outperformed ACA in steady-state performance and scaling responsiveness; ACA provided faster cold-start activation. Non-parametric statistical tests (Mann–Whitney U and Kruskal–Wallis with Bonferroni-corrected post hoc comparisons), together with effect-size analysis (Cliff’s δ and Cohen’s d), support the main conclusions of the study. Statistically significant differences are observed for most load levels, whereas intermediate concurrency levels show non-significant differences between ACA and Koyeb. We also introduce the deployment topology-aware benchmarking model, which decomposes client-observed end-to-end latency into topology-dependent and topology-independent components; the associated topology contribution ratio (TCR) – ranging from TCR ≈ 0 for local loopback to TCR ≈ 0.37 for ACA at C = 10 – enables principled cross-environment comparison without conflating network-path effects with platform processing performance.