Embodiments described herein rapidly pre-migrate child virtual machines (VMs) by leveraging shared memory resources between parent and child VMs. In a first, proactive phase, at least a portion of the memory of a parent VM is migrated to a plurality of potential target hosts. In a second, reactive phase, after a request is detected to migrate a child VM to a selected target host, differences in memory are transferred to the selected target host. Aspects of the disclosure offer a high performance, resource efficient solution that outperforms traditional approaches in areas of software compatibility, stability, quality of service control, resource utilization, and more.