Kubernetes

Docker Swarm

A separate modular design project that has its own dependencies.

Native container orchestration solution available out of the box.

Relatively steep learning curve due to new concepts and complex architecture.

Easy to get started; uses familiar terminology; more lightweight.

A pod is a minimal unit of deployment which represents a group of containers. Integration with other applications is accomplished via services that in this case represent a consistent IP:port pair.

Application deployed in containers as services across an entire cluster or a subset of workers using labels.

Auto-scaling is supported via deployments/replication controllers by specifying a desired number of pods. Dynamic auto-scaling that takes CPU utilization into account is provided by the HorizontalPodAutoscaler resource.

Auto-scaling is not supported out of the box; manual scaling is still possible.

A persistence storage layer is separated into two components, PVs and PVCs, which are dynamically bound together on request and can be used to implement shared storage.

Storage volumes are mounted directly into containers.

New masters can join an existing cluster, but promotion/demotion of a node is not supported.

Worker nodes can be easily promoted to managers and vice versa.

Services are assigned unique DNS names based on the projects they were created in and their names, so each service can reach any other in the same namespace by using its name without domains.

Each service is registered in an internal DNS with the name based solely on the name of the service itself.