VM Migration - Cold Hot and Live
Definition
VM migration moves a virtual machine from one physical host/hypervisor to another. The lecture motivates it through consolidation, maintenance, and availability/disaster-recovery scenarios.
Three types in the professor’s terminology
| Type | Execution handling | State requirement |
|---|---|---|
| Cold | Shut down, move, restart | Persistent disks/configuration; previous running state need not survive |
| Hot | Pause, move, resume | Preserve RAM and relevant CPU/device/hypervisor state |
| Live | Copy while running, then hand off | Preserve execution state and account for pages modified during copying |
“Hot” here means the slide’s pause-move-resume sequence; retain that terminology when comparing the lecture’s three categories.
What state belongs to a VM?
PDF page 103 / printed slide 109 distinguishes:
- Virtual hardware configuration, including NIC type and MAC address.
- Boot image, modified OS disk, and data disk.
- Guest RAM: application variables, processes, OS state, buffers, open-file/socket state.
- CPU and emulated-device state maintained by the hypervisor.
- Hypervisor-side structures used to execute the VM.
- External network delivery to its IP/MAC identity.
Study clarification: these categories identify state to preserve, transfer, share, or recreate. They do not imply every implementation blindly copies every cache/TLB structure byte-for-byte. The boot-image/live-disk separation is the diagram’s model, not a universal storage layout.
Live-copy difficulty
While RAM is copied, the guest can modify previously copied pages. These dirty pages require further handling before the final consistent handoff.
Study example: if a guest modifies memory faster than the available link can move the changing state, a naive repeated-copy process may struggle to finish. “Live” emphasizes continued operation during most of migration, not a universal zero-pause guarantee.
Storage and networking
Shared/remote storage can avoid transferring the full live disk between hosts. Network reachability still has to follow the VM; the lecture mentions gratuitous ARP and bridge-versus-NAT differences. Physical device assignment can complicate preserving or recreating device state.
Common mistakes
- Copying only a disk image and expecting the current running session to survive.
- Assuming the IP must change in every migration, or can remain reachable without network work.
- Assuming migration itself provides a complete high-availability or recovery strategy.
Related: Virtualization Agility and VM Migration, QEMU and KVM, I-O Virtualization - Emulation PV and Passthrough, Libvirt virsh and virt-manager.
Source
Lecture 03 PDF pages 101-107 (printed slides 107-113). Context: Lecture 03 - Computing Virtualization Technologies and Tools. State qualifications, convergence example, and common mistakes are study explanations.