Goal

• As an owner of a VM, I would like my guest to suffer no more than 50ms time skew during live migration, so I can run timing-sensitive applications in it.

Acceptance Criteria

• report the average and standard deviation of guest time skew during pre-copy live migration.

Details

• Prepare two bare-metal nodes with chronyd so that the time difference between them is below 1ms.
• Start a RHEL9 guest on one node.
• Ensure that chronyd or any other time-manipulating service is not running in the guest.
• repeat 100 times in order to measure the migration-driven time skew |pre-migration-diff - post-migration-diff|
  • Measure the time difference between the guest and and its node in 1ms precision (or better).
  • Migrate the guest. Measure the time difference again.

• Publish the average and standard deviation of our time skew.
• Ensure that it does not become worse in the future