Kubernetes Core Concepts, Comparison with Docker Swarm, and Practical Deployment Tips

The article begins with an overview of the rapid growth of container technologies and the emergence of Kubernetes (k8s) as the dominant container orchestration platform, emphasizing its impact beyond simple container management.

Core Concepts : It explains that a pod is the smallest deployable unit in k8s, capable of running one or multiple containers that share network and storage. Controllers such as Deployment , StatefulSet , and DaemonSet manage pod lifecycles, while Service objects expose pods via different types – ClusterIP (iptables NAT), NodePort, LoadBalancer, and Ingress (7‑layer HTTP load balancing).

Technical Comparison : The article contrasts k8s with Docker Swarm, highlighting differences in scheduling granularity (pod vs container), auto‑scaling (HPA in k8s), load‑balancing options, rolling updates, and disaster recovery capabilities. It notes that Swarm offers faster container startup but lacks the fine‑grained management and robustness of k8s.

Practical Experience : The author describes using the open‑source Wayne dashboard to simplify cluster management, reduce YAML learning overhead, and provide role‑based access control, while also acknowledging the added complexity for small teams.

Deployment Considerations : Recommendations include using Alibaba Cloud mirrors for yum, aligning Docker’s cgroup driver with kubelet, switching kube‑proxy mode from iptables to ipvs for performance, and editing the kube-proxy ConfigMap to set mode: ipvs and scheduler: round‑robin .

Performance Test Data : Results from a self‑built k8s cluster (60 and 40 concurrent threads) are presented alongside bare‑metal Java benchmarks, showing k8s TPS around 4.5k with ~12 ms response time versus Java TPS around 7.3k with ~7 ms response time, illustrating the modest overhead of container orchestration.

Overall, the article concludes that while k8s introduces some performance penalty compared to bare‑metal, its advantages in automated deployment, resource control, and ecosystem support outweigh the costs for most production services.