High availability and live migration for virtual machines
In short: Clusters of hypervisor hosts with shared storage let VMs restart automatically when a host fails and move between hosts while running. Live migration enables maintenance with no downtime, load balancing spreads work, and admission control reserves enough spare capacity to survive a failure.
Why clusters matter
When dozens of VMs run on one physical host, that host becomes a big single point of failure. Hypervisor platforms solve this by grouping hosts into a cluster that shares storage and networks. The cluster software monitors every host and can move or restart VMs. Examples include vSphere HA and DRS, Windows Server Failover Clustering with Hyper-V, and Proxmox VE HA.
HA restart
With VM high availability (HA), hosts exchange heartbeats. If one host stops responding, the others confirm the failure (often by checking the shared storage as well, to avoid acting on a mere network problem) and restart its VMs elsewhere. This is not zero downtime: the VMs reboot, so expect a few minutes of outage. It does, however, remove the need for someone to notice and act at 3 a.m.
Admission control makes HA work in practice. It reserves enough spare CPU and memory, for example one host's worth (N+1), so there is room to restart failed VMs. Without it, a busy cluster may have nowhere to put them. Set restart priorities so infrastructure such as domain controllers and DNS comes up before application servers.
Live migration
Live migration (vMotion in VMware, Live Migration in Hyper-V) moves a running VM to another host without shutting it down. Memory is copied in rounds while the VM keeps working; then the VM is paused very briefly, the last changes are sent, and it resumes on the new host. The new host announces the VM's MAC address so switches update their tables. Users typically notice nothing.
Requirements usually include:
- Shared storage, or a storage migration at the same time (shared-nothing migration).
- The same port groups and VLANs on both hosts.
- Compatible CPUs; features such as VMware EVC or Hyper-V processor compatibility mode hide differences between CPU generations.
- A fast, isolated migration network, ideally 10 Gb/s or more, encrypted if it crosses untrusted links.
Other availability features
| Feature | What it does | Downtime on host failure |
|---|---|---|
| HA restart | Reboots VMs on surviving hosts | Minutes (VM restarts) |
| Fault tolerance | Keeps a live shadow copy of a VM on another host | Near zero, but costly and limited |
| Load balancing (DRS-style) | Moves VMs to spread load automatically | None; prevents hotspots |
| Maintenance mode | Live migrates all VMs off a host before patching | None for planned work |
| Application clustering | Database or app replicas with their own failover | Seconds, app-aware |
Affinity and anti-affinity rules control placement. Anti-affinity keeps two domain controllers or two web servers on different hosts, so one failure cannot take out both.
Tip: HA protects against host hardware failure, not against a corrupted VM, a bad patch or the loss of the whole site. Combine it with backups, application-level redundancy and disaster recovery to another location.
Test HA regularly: put a host into maintenance mode, or in a test cluster pull its power, and check that VMs restart in the right order and that admission control left enough room.
Practice
Does VM HA give zero downtime when a host fails?
What is the purpose of admission control?
Why would you set an anti-affinity rule for two domain controllers?
Key takeaways
- Clusters with shared storage let VMs restart or move between hosts.
- HA restarts VMs after host failure; expect a short outage.
- Live migration moves running VMs with no noticeable downtime.
- Admission control and restart priorities make HA reliable.
- Use anti-affinity for redundant pairs and combine HA with backups and DR.
Original content written by Mohamed Fayas for FM TecHub. © 2026 FM TecHub. You're welcome to share a link to this page; please don't copy the text elsewhere without permission.