Nobody notices live migration while it works. When it breaks, it usually shows up in the middle of a maintenance window, exactly when you need to drain a host. Below we sort the cases where live migration fails or takes minutes by symptom; for each, we give the first place to look in the console, the likely causes and the permanent fix. The audience is system administrators running a working Hyper-V environment.
The stages of a live migration
A diagnosis made without knowing at which stage the error occurs is a guess. The process has four steps:
- Setup: The source host connects to the destination, authentication takes place, and the VM’s configuration and an empty “skeleton” VM are created on the destination. Memory is reserved on the destination.
- Memory copy: While the VM keeps running, memory pages are copied to the destination. Pages that change (get dirty) in the meantime are tracked and sent again; the iterations continue until the remaining set of changes is small.
- State transfer: The VM is paused on the source for a very short time, and the last dirty pages and processor/device state are transferred.
- Handover: Storage access moves to the destination, the VM resumes on the destination, and on the network side the switches are notified of the MAC address’s new location.
In a Failover Cluster, since VM files sit on shared storage (CSV or SMB), only memory and state are moved. In a non-clustered “shared nothing” migration, virtual disks are also copied over the network; the time grows in proportion to disk size. Before starting to troubleshoot, be clear about which scenario you are in within the system infrastructure. The two paths have different authentication and network requirements.
First diagnosis by symptom
The first job is to find at which stage the error occurs. Event logs and the Compare-VM pre-check are used together; which channel to look at and why is below.
| Channel | What to look for |
|---|---|
| Microsoft-Windows-Hyper-V-VMMS-Admin | Migration start, authentication and configuration errors |
| Microsoft-Windows-Hyper-V-Worker-Admin | VM worker process, processor and memory errors |
| Microsoft-Windows-Hyper-V-High-Availability-Admin | Migration errors through the cluster (e.g. event 21502) |
| Microsoft-Windows-FailoverClustering/Operational | Cluster network, resource and node events |
# Errors and warnings in Hyper-V channels in the last 2 hours
$start = (Get-Date).AddHours(-2)
Get-WinEvent -FilterHashtable @{
LogName = 'Microsoft-Windows-Hyper-V-VMMS-Admin',
'Microsoft-Windows-Hyper-V-High-Availability-Admin',
'Microsoft-Windows-Hyper-V-Worker-Admin'
Level = 2,3; StartTime = $start } |
Select-Object TimeCreated, LogName, Id, Message | Format-List
# Cluster log (last 30 minutes, local time)
Get-ClusterLog -Destination C:\Temp -TimeSpan 30 -UseLocalTime
Compare-VM reports what would block the VM on the destination host without actually starting the migration. For non-clustered migrations, this is the first command to run:
$r = Compare-VM -Name APP01 -DestinationHost HV02
$r.Incompatibilities | Format-Table MessageId, Message -AutoSize
The output clearly lists causes such as processor incompatibility, a missing virtual switch, the configuration version or an unreachable file path. In a cluster, the Hyper-V configuration tests in the “Validate Cluster” report provide a similar pre-check. Don’t retry without reading the report.
Processor compatibility
A running VM sees and uses the instruction set features of the processor on the host where it was started. If the destination host’s processor is an older generation, a feature the VM uses may not exist on the destination; the migration is rejected with an error meaning “the VM is using processor-specific features not supported on the destination computer.” This mostly shows up after a newer-generation server is added to the cluster.
The fix is to enable processor compatibility mode for the VM. The mode reduces the features visible to the VM to a common subset and is set while the VM is powered off:
Stop-VM -Name APP01
Set-VMProcessor -VMName APP01 -CompatibilityForMigrationEnabled $true
Start-VM -Name APP01
Points to watch:
- Compatibility mode works only between processors from the same vendor; live migration between Intel and AMD hosts isn’t possible. In that case the VM is moved powered off.
- In workloads that make heavy use of new instruction set extensions (e.g. vector operations), compatibility mode can affect performance. Instead of applying the setting to all VMs in bulk, use it on VMs that need to move between hosts of different generations.
- Windows Server 2025 and Azure Local have a new processor compatibility mode that dynamically calculates the common feature set of the nodes in the cluster. Verify the parameters for your environment’s version in the Microsoft documentation before using it.
A blocker from the same family is the VM configuration version. A VM with a configuration version upgraded on a newer Windows Server host can’t move to an older host that doesn’t support that version. You can see the supported versions with Get-VMHostSupportedVersion. In mixed-version clusters (rolling upgrade), don’t run Update-VMVersion until all nodes are upgraded. There’s no going back.
Authentication: Kerberos and CredSSP
In non-clustered live migrations (including shared nothing), the source host must authenticate to the destination. Hyper-V offers two methods:
| Method | How it works | Typical error |
|---|---|---|
| CredSSP | The user’s credentials are passed from the source host to the destination | Fails when the migration is started remotely from a console |
| Kerberos | Constrained delegation is used for computer accounts | “Access denied” or an authentication error if delegation is missing |
Migration with CredSSP works only when you are logged on directly to the source host (console or RDP). A migration started from Hyper-V Manager or Windows Admin Center on a management workstation fails because of the “double hop” problem. On hosts with Credential Guard enabled, passing credentials with CredSSP is also blocked. The permanent fix for remotely managed environments is Kerberos:
- In Active Directory, on each Hyper-V host’s computer object, on the “Delegation” tab, select “Trust this computer for delegation to specified services only” and “Use Kerberos only.”
- Add the
cifsandMicrosoft Virtual System Migration Serviceservices for every other host. - Set the authentication type to Kerberos on each host.
Enable-VMMigration
Set-VMHost -VirtualMachineMigrationAuthenticationType Kerberos
Get-VMHost | Select-Object VirtualMachineMigrationEnabled,
VirtualMachineMigrationAuthenticationType, VirtualMachineMigrationPerformanceOption
For the delegation change to take effect, the Kerberos tickets need to be renewed. If you get an error on the first try after the change, restarting the hosts one by one or clearing the computer account’s ticket cache with klist purge -li 0x3e7 solves it. Migrations started through the Failover Cluster don’t need this delegation; the cluster uses its own communication.
Network and virtual switch mismatches
A common configuration error is that the virtual switch name the VM network adapter is connected to doesn’t exist on the destination host. Hyper-V matches by switch name; “SETswitch” and “SET-Switch” are different switches. A single hyphen stops the migration. In a cluster, switch names must be exactly the same on all nodes. In a non-clustered migration, you can complete the migration by connecting the adapter to the destination switch through the Compare-VM report:
$r = Compare-VM -Name APP01 -DestinationHost HV02
$r.Incompatibilities[0].Source | Connect-VMNetworkAdapter -SwitchName "SETswitch"
Compare-VM -CompatibilityReport $r # check again
Move-VM -CompatibilityReport $r
Other network-related causes:
- Migration network selection: If the subnet defined with
Add-VMMigrationNetworkon a non-clustered host can’t be reached from the destination, the migration times out. In a cluster, the networks selected in “Live Migration Settings” in Failover Cluster Manager are used; the management network sitting at the top of the list is a classic cause of slow migrations. - Firewall: TCP 6600 for live migration, and TCP 445 with the SMB performance option, must be open in both directions.
- MTU inconsistency: If jumbo frames are on at the host but off at a switch in between, setup with small packets succeeds, and the migration hangs or drops during the large memory transfer. Verify end to end with
ping -f -l 8972. - Storage path: In a cluster, an ISO attached to the VM from the host’s local disk, a checkpoint written to local disk or a Smart Paging file blocks migration. Check that all file paths in the VM settings are on a CSV or shared SMB. A forgotten ISO is the most common.
- Insufficient memory on the destination: If the destination host can’t reserve the VM’s Startup memory, the migration is rejected at the setup stage.
To build the host network right from the start, see the Hyper-V SET virtual network design article.
The migration finishes but is very slow
If the migration completes successfully but takes very long, the problem is usually the balance between bandwidth and the VM’s memory change rate. A VM that writes heavily all the time can’t finish the copy iterations if the link is insufficient. Bandwidth isn’t enough. The iterations drag on forever.
| Option | Behavior | When it fits |
|---|---|---|
| TCP/IP | Memory directly over TCP | Troubleshooting, very simple environments |
| Compression (default) | Memory is compressed with the host CPU | Links below 10 GbE, hosts with idle CPU |
| SMB | SMB Multichannel and SMB Direct (RDMA) if available | 10 GbE and above, networks with RDMA |
Set-VMHost -VirtualMachineMigrationPerformanceOption SMB
Set-VMHost -MaximumVirtualMachineMigrations 2 -MaximumStorageMigrations 2
Set-VMHost -UseAnyNetworkForMigration $false
Add-VMMigrationNetwork 10.0.30.0/24
Increasing the number of simultaneous migrations doesn’t always shorten the total drain time; since bandwidth is shared, individual migrations get longer and busy VMs may not complete. Measure the time to drain a host before the change and repeat the same measurement after the change. A setting made without measuring isn’t a setting.
Error–cause–fix table
| Symptom | Likely cause | Fix |
|---|---|---|
| Processor-specific feature error | Different CPU generation | Processor compatibility mode |
| Authentication error when started remotely | CredSSP, double hop | Kerberos and constrained delegation |
| Virtual switch not found | Different switch name | Same name on all hosts |
| Configuration version not supported | Migration to an older host | Postpone the version upgrade |
| File can’t be accessed | Local ISO, checkpoint path | Move the paths to shared storage |
| Migration very slow | Wrong network, low bandwidth | Set the migration network and the SMB option |
| Drop during large transfers | MTU inconsistency | End-to-end MTU test |
A practical way to tie these checks to scheduled maintenance is to move a test VM from each host to all nodes before the maintenance window; the problem surfaces the day before, not on maintenance night. For regular maintenance and troubleshooting support, see our technical support and scheduled maintenance page, and for an architecture review, our virtualization and data management page. Prerequisites for cluster design are in the Hyper-V failover cluster planning article; Microsoft’s host preparation steps for non-clustered migration are in the Microsoft Learn documentation.
Frequently asked questions
Can live migration be done between hosts with Intel and AMD processors?
No. Processor compatibility mode works only between different processor generations from the same vendor. Between Intel and AMD hosts, the VM must be shut down and moved (cold migration).
Is Kerberos constrained delegation also needed in a cluster?
Not for migrations started with Failover Cluster Manager or the cluster cmdlets; the cluster uses its own communication. It is needed for migrations between non-clustered hosts (including shared nothing) if Kerberos is selected as the authentication type. It is also the way to start a migration remotely, without logging on to the source host.
Which ports must be open for live migration?
Live migration uses TCP 6600 by default. If the SMB performance option is selected, TCP 445 is also needed; if the VM files sit on an SMB share, access to 445 is required too. Opening rules with the built-in Hyper-V rules in Windows Firewall produces fewer errors than writing port rules by hand.
Which is faster, the Compression or the SMB option?
On links below 10 GbE, if the host has spare processor capacity, Compression usually gives better results. On 10 GbE and above, especially RDMA-capable networks, the SMB option is faster thanks to SMB Multichannel and SMB Direct and reduces CPU load. Before deciding, measure the time to drain a host in your own environment.