Live migration of virtual machines among edge networks via WAN links

This paper addresses the potential impact of emerging technologies, like software defined networking (SDN) and network virtualization (NV), on future network evolution. It is argued that the above mentioned technologies could bring a significant disruption at the edge networks, where it will be possible to develop distributed clouds of virtual resources running on standard hardware. This paper deals with a key technical challenge behind this vision: the capability of dynamically moving virtual machines (VMs), which run network services, functions, and users applications, among edge networks across wide area interconnections. Specifically, we focus on the problem of live migrating the memory state of a single VM between two physical machines, which are located at the edge and are inter-connected by a wide area network (WAN). With reference to a pre-copy mechanism, aimed at iteratively transferring the memory content to the destination machine, we develop a simplified mathematical model that unveils the dependence of the total migration time and downtime of the VM memory transfer on the mainWAN parameters, such as capacity, buffering, and propagation delay. Numerical results obtained by Matlab simulations are presented to demonstrate the feasibility of live VM migration across WANs and the directions of RT&D activities.