Poor performance during live migration of a Hyper-V VM with standard checkpoints

When you live migrate Hyper-V virtual machines (VM) that use standard checkpoints, the migration performance is poor. The migration may fail because of a time-out period expiration.

Standard checkpoints capture the state, data, and hardware configuration of a running VM. They are intended for use in development and test scenarios. Because standard checkpoints contain lots of data, such as the virtual machine’s memory state, live migrating VMs that use standard checkpoints is more resource-intensive than live migration that uses production checkpoints. Depending on the underlying infrastructure, this can cause live migration failure because of networking time-outs.

In some scenarios, storage performance for the system drive can become a bottleneck. During live migration, Hyper-V uses locations on the system drive to create temporary files as it creates the virtual machine. If the destination node can no longer write the data to disk in a reasonable time, the source node may time out while it waits for the data transfer to continue.

To avoid this issue, follow these steps:

1. Use production checkpoints instead of standard checkpoints.  Production checkpoints don't affect live migration. We recommend that you use production checkpoints in production environments. Standard checkpoints are intended for use in development and test scenarios.
2. Use faster system drives to improve performance if the system drive is the data transfer bottleneck.
3. Limit the number of simultaneous live migrations. Because simultaneous live migrations can drain a node faster, performing too many migrations can significantly affect live migration performance. By default, Hyper-V is configured to allow two simultaneous live migrations. You can change this value by running the following PowerShell cmdlet (for example):

Set-VMHost -MaximumVirtualMachineMigrations 2

For more discussion about how to tune a live migration when you use a hyperconverged infrastructure, see this blog post.