In a virtualized cluster, Live VM Migration is the process of moving a running virtual machine from one physical host to another without shutting it down. This technique is widely used for:
- Load balancing
- High availability
- Server consolidation
- Disaster recovery
Unlike physical-to-physical failover, virtual machines can migrate between hosts without affecting service continuity — even if a VM’s host fails.
States of a Virtual Machine (VM)
A VM can exist in the following four states:
- Inactive: VM is not enabled.
- Active: VM is running and performing tasks.
- Paused: VM is temporarily halted but still consuming resources.
- Suspended: VM state is saved to disk and halted completely.
Steps of Live Migration of a VM
Live migration involves 6 steps:
Step 0 & 1: Initiate Migration
- Identify the VM to migrate and the destination host.
- This may be done manually or automatically based on strategies like load balancing.
Step 2: Memory Pre-Copy
- Copy the entire memory of the VM to the destination host.
- The dirty pages (memory pages modified during transfer) are copied again in multiple rounds.
- This continues until the dirty data becomes small enough for final transfer.
- During this, the VM continues to run with minimal impact.
Step 3: Suspend VM and Copy Final Data
- The VM is paused to transfer the remaining dirty pages, CPU state, and device/network info.
- This introduces a short downtime, where the VM service is temporarily unavailable.
- This downtime must be very short (e.g., milliseconds) for a seamless experience.
Step 4 & 5: Commit and Activate on New Host
- VM is activated on the destination host using the transferred state.
- Network connections are re-established, and all dependencies on the source host are cleared.
- The old VM is deleted from the source.
Performance Effects of Live Migration
An experiment with 512 KB file transfers to 100 clients shows:
- Initial throughput: 870 MB/s
- During precopy: 765 MB/s (for 63s)
- Later iterations: 694 MB/s (for 9.8s)
- Downtime: Just 165 milliseconds
This indicates that live migration causes very low performance overhead, ensuring that services remain almost uninterrupted, which is vital for cloud and enterprise systems.