The match everyone asks about
Docker Swarm is dead, they've said since 2019. Yet every week we see customers happily deploying Swarm clusters — and single-node Kubernetes clusters consuming 2 GB of RAM to run three containers. The real question isn't fashion, it's fit to scale.
Swarm: orchestration without the tax
docker swarm init and you have an orchestrator: replicated services, rolling updates, secrets, overlay network. The learning curve is measured in hours, not weeks. On one to five VPS, Swarm does everything 95% of teams need with zero additional components.
Kubernetes: powerful, but demanding
K8s shines beyond 5-10 nodes with dedicated teams. Below that, you pay the complexity (etcd, control plane, manifests) without reaping the benefits. k3s softens the bill — 512 MB of RAM — but it's still Kubernetes to administer.
Our decision table
1 VPS, a few services: Compose. 2-5 VPS, rolling deploys: Swarm. 5+ nodes, platform team, fine autoscaling needs: k3s/K8s. Whatever the orchestrator, node performance remains the foundation: NVMe Gen4 for images and volumes, fast inter-site networking (10 Gbps internal with us) for the overlay.
The frequent hybrid case
Many of our customers pair: a production VPS on Swarm, plus a small VPS dedicated to tooling (registry, monitoring, backups) — simplicity without the single point of fragility.