Coordinated Live Migration of Virtual Machines

This paper proposes a hypervisor-level live migration mechanism called coordinated live migration of Virtual Machines (VMs), which enables the concurrent migration of multiple VMs running parallel applications between host computers. Coordinated live migration is a collective operation that performs live migration on multiple VMs uniformly. It is also highly transparent because it operates at the hypervisor level. Our solution comprises a software framework for managing VMs and a novel coordinated live migration protocol. We introduce a live migration controller, a software agent responsible for orchestrating the protocol. In our design, the coordinated live migration protocol collaborates with local live migration mechanisms to transfer the state of each migrating VM to its destination. In this paper, we develop two local live migration mechanisms: the extended precopy and the memory-bound precopy approaches. We have implemented prototypes of coordinated live migration using both mechanisms and conducted preliminary experiments by performing coordinated live migrations on VMs running an MPI application to assess their performance. Experimental results demonstrate that the coordinated live migration protocols using both the extended and memory-bound precopy mechanisms perform successfully. There is a tradeoff between using the protocol with the extended precopy and memory-bound precopy mechanisms; nevertheless, we recommend the memory-bound approach, as it requires fewer parameter settings.