Análise de Desempenho entre Docker e KVM: Um Estudo de Caso em Microsserviços Web sob Restrição de Recursos

Abstract

This work presents a comparative performance analysis between Docker container technology and KVM hypervisor-based virtualization, focusing on the execution of web microservices under hardware resource constraints of 1 vCPU and 1 GB of RAM. The study is motivated by the relevance of low-cost VPS environments and edge computing scenarios, in which the overhead imposed by the virtualization layer can directly affect the application service capacity. For this purpose, a RESTful API was developed in Node.js v22.22.2 with Express.js, specifically implemented for this study and exposing dedicated endpoints designed to stress different computational resources. The application was submitted to four experimental scenarios: baseline and cold start, network stress, CPU saturation, and memory pressure, using Grafana k6 as the load testing tool. Each scenario was executed with 15 independent repetitions per evaluated configuration, allowing the analysis of metrics such as latency, throughput, CPU utilization, memory consumption, and startup time. The results show that Docker presents a significant advantage in cold start, being 5.2 times faster than KVM. Under moderate load, both technologies presented similar performance in latency and throughput. Under more intensive workloads, KVM demonstrated better behavior in tail latency, throughput under CPU saturation, and performance under memory pressure, although the latter scenario must be interpreted considering the experimental configuration adopted. The findings indicate that the historical overhead of KVM has been reduced in modern environments, making it competitive for stable transactional workloads, while Docker remains superior in scenarios requiring fast startup and higher instance density.