Consistent migration of a group of virtual machines using source and destination group messaging

摘要

A plurality of virtual machines (VMs) is migrated from a source group to a destination group in such as way as to achieve consistency and either availability or group preservation. Execution of VMs in the source group is selectively delayed during state migration so that memory transfer of all the VMs in the group will converge roughly at the same time. After VM state transfer to the destination group, execution switch-over is coordinated using different handshake and acknowledgement messages, passed either through a “leader” VM in each group, or directly between source-destination VM pairs.

主权项

1. A method for migrating a group of virtual machines (VMs) from at least one source platform to at least one destination platform, comprising:
pre-copying contents of memory associated with each source VM to memory associated with a corresponding respective one of the destination VMs such that copying the memories of all source VMs to be migrated is completed within a pre-copy convergence time window; stunning each source VM after completion of the pre-copying contents of the memory associated with it; checkpointing the stunned source VM and transferring its checkpoint to the destination platform; transmitting from the destination platform to the source platform a handshake message indicating readiness to begin execution of at least one of the destination VMs on the destination platform; receiving from the source platform an acknowledgement message indicating readiness to transfer execution from at least one source VM to the corresponding destination VM; starting execution of each destination VM for which the acknowledgement message has been received; and switching execution from the VMs running on the source platform to the corresponding VMs on the destination platform by:
selecting a source leader VM and a destination leader VM;when the destination leader VM has received destination VM resumption messages from all the other destination VMs, transmitting to the source leader VM a destination group resumption message, which the source leader VM forwards to the source VMs;when the source leader VM has received source VM acknowledgement messages from all source VMs, which each power off after sending the respective acknowledgement message, sending a source group acknowledgement message to the destination leader VM; andforwarding the source group acknowledgement message to each destination VM, whereupon the destination VMs resume execution.