The foundation behind Nokkela.

nokkela.host offers both, because they solve different problems: a virtual machine gives you a full, isolated operating system on KVM and Proxmox VE with flexible CPU, RAM and disk, while a container packages one application so it stays reproducible and portable across Docker and Kubernetes. Legacy or stateful systems usually go into VMs, modern CI/CD workloads into containers. Dedicated VMs and whole dedicated clusters are both available.

CEPH block storage on nokkela.host keeps every block replicated across several nodes, so the loss of a single disk or node does not take the volume offline. The cluster detects the missing copies and rebuilds them onto the remaining nodes on its own – that is the self-healing property. Capacity scales linearly by adding nodes, and thin provisioning means volumes only occupy what they actually use.

No – snapshots on nokkela.host are a fast rollback point inside the same CEPH cluster, not an independent backup. They are ideal before an update or a migration, because a volume can be returned to its previous state quickly, but they share the fate of the cluster they live in. We therefore combine snapshots with a separate backup, which we plan with you as part of the platform design.

nokkela.host provides GPUs in both forms. With PCIe GPU passthrough a virtual machine gets a physical card to itself, which suits training runs and large models that need the full card and its memory. With vGPU the same card is shared between several VMs or containers, which usually fits inference and development better. The AI workloads of nokkela.ai run on exactly this platform.

nokkela.host uses live migration, so a running virtual machine is moved to another node in the cluster before the host is taken out of service. Because the VM's disks live on shared CEPH storage rather than on the host, nothing has to be copied over. The cluster spans several nodes and sites in Germany and Finland, and HA restarts a VM elsewhere if a node fails unexpectedly.

Yes – nokkela.host runs Docker and Kubernetes as first-class workloads, and container platforms usually sit on VMs of the cluster, which keeps orchestration separate from the hypervisor layer. Isolation works on two levels: network segmentation and resource limits per container, so one workload does not starve its neighbours. The setups are CI/CD-ready, and the agent runtimes of nokkela.io use the same model.

nokkela.host is covered by monitoring, logging and alerting across the whole storage and compute stack, so capacity trends and faults become visible early. Hosts are hardened, networks segmented and traffic encrypted, and the CEPH cluster reports its own replication state. A public status page for the Nokkela services is linked from every page in the group, and the people who answer an alert are the same ones who operate the platform.