Publication details
- Journal: IEEE Xplore Digital Library, p. 1–6, 2026
-
International Standard Numbers:
- Electronic: 2473-2001
- Links:
Microservice-based architectures are quickly replacing monolithic software systems, with Docker and Kubernetes being the defacto platforms for deploying and orchestrating these distributed services. Yet, the resilience, scalability, and operational robustness of microservices under real-world stress conditions remain as open questions. This paper presents a systematic evaluation of two popular microservice benchmarks—Boutique and Sock Shop—subjected to high-load conditions and controlled chaos experiments, including pod failures, network delays, and network partitions. We containerize both applications and assess their performance across multiple Kubernetes distributions (K0s-K3s), Container Network Interfaces (CNI), and horizontal and vertical pod autoscaling (HPA, VPA, Hybrid) profiles. Nearly 300 experiments were executed over approximately ∼150 hours spanning 4 load profiles and 5 runs for each configuration. Our initial results show that Sock Shop is a challenging, stateful microservice benchmark that is hard to scale, but K3s-Calico CNI combination provides a stable baseline until high loads. We find that microservice-based applications must be carefully managed based on performance, scalability, and reliability objectives.