Performance, Cost, and Sustainability Trade-offs in E-Commerce Platform Architectures under Peak Load Conditions

Abstract

As e-commerce continues to grow, platforms face the challenge of delivering high performance—especially during peak traffic events such as flash sales or festival days—while also maintaining cost efficiency and environmental sustainability. This study presents a comprehensive benchmarking of four architecture styles—monolithic, microservices (REST), event-driven microservices, and serverless—evaluated under realistic load scenarios. Key metrics collected include latency percentiles (P50, P90, P95, P99), throughput (requests per second), error rates, infrastructure cost per transaction, and energy consumption (or proxy of power use).

Recent studies (for example “Reducing Environmental Impact with Sustainable Serverless Computing” (Akour et al., 2025)) show that serverless computing can reduce energy consumption by up to 70% and operational costs by up to 60%, but also point out that these benefits are heavily dependent on workload type and configuration (caching, cold starts etc.). MDPI Additionally, cold-start latency remains a critical challenge in many function-as-a-service (FaaS) environments. Systematic reviews find that cold starts can significantly degrade user experience, especially under sustained peak load, and are influenced by factors such as runtime, deployment region, function size, and trigger type.

In our experiments, we observe that while serverless architectures offer strong scaling and cost benefits under moderate load bursts, their performance degrades under heavy sustained load, exhibiting substantially higher cold-start latency and disproportionately increasing energy costs in worst-case percentiles. On the other hand, event-driven microservices, when properly configured, provide a more stable balance between performance, scaling cost, and energy usage, with lower tail latency compared to serverless in most peak scenarios. Based on these findings, we propose configuration best practices including: proactive cold-start mitigation (warm-pooling, library optimization), intelligent autoscaling thresholds, efficient caching strategies, and judicious use of event brokers.