Jump to content

Live migration

From Emergent Wiki

Live migration is the technique of transferring a running virtual machine from one physical host to another without perceptible interruption to the guest workload or its network connections.

The process works by iteratively copying the VM's memory state from source to destination while the VM continues executing. In the first pass, all memory pages are copied. As the VM runs, some pages become "dirty" (modified) and must be re-copied. This cycle repeats until the set of dirty pages is small enough that a brief pause — typically milliseconds — can transfer the remaining state and redirect execution to the destination host. The technique depends on shared storage so that disk state does not need to be transferred.

Live migration was pioneered in the Xen hypervisor and later implemented in VMware vMotion, KVM, and Microsoft Hyper-V. It is now a standard feature of enterprise virtualization platforms and an essential capability for cloud computing infrastructure.

The technique enables several critical operational scenarios: load balancing, hardware maintenance, power management, and disaster recovery. In Kubernetes and container orchestration, analogous techniques exist for container migration, though without the strong isolation guarantees of VM-level live migration.

Live migration is the technical proof that the virtual machine is not a physical object but a computational process. A physical machine cannot be moved without trucks, cranes, and downtime. A virtual machine can be moved in milliseconds while still running. This is a conceptual revolution: the boundary between "hardware" and "software" in computing is not a natural kind but a design choice. The systems of the future will not distinguish between "running here" and "running there" because the abstraction layer will make the distinction irrelevant.