Comparing performance and scalability of a microservice oriented architecture vs a monolithic architecture for e‐commerce platforms

Abstract

This dissertation presents a comprehensive performance evaluation and architectural comparison between microservices and monolithic systems within the context of an e‐commerce application. The study involves the development of two implementations: one as a distributed system comprising several microservices, namely the user‐service, order‐service, and product‐service, coordinated by an Application Programming Interface (API) gateway that converts JSON Web Token (JWT) tokens into custom headers for inter‐service communication, and the other as a traditional monolithic system. The microservices integrate Kafka for asynchronous messaging, and both architectures integrate PostgreSQL for shared data persistence, with container orchestration provided via Docker. Performance tests using Apache JMeter reveal that, under identical resource allocations, the monolithic architecture consistently delivers lower latency and higher throughput due to the absence of inter‐service overhead. Only when compute cores are reallocated to the hot path in the microservices setup does it surpass the monolith, achieving a 37% reduction in mean latency and a 59% increase in throughput, albeit at the expense of nearly double the peak CPU utilisation. In addition, database statistics gathered and visualised using Postgres‐exporter, cAdvisor and Grafana provide insights into query patterns and resource utilisation, confirming that neither architecture is database bound and that application and network factors dominate end to end performance.