Power-efficient virtual machine replication in data centers

By replicating virtual machines (VMs) and placing replica copies into data centers, not only does it distribute the requests to virtual machines into different physical machines, thus reducing server load, but also it achieves fault tolerance in risks of server failures by placing multiple copies of a VM on different servers. This paper studies the virtual machine replication problem (VMR) in data centers, with the goal of minimizing the total power consumption in this process. To guarantee that each VM is available in the event of server failure, it replicates R copies of each VM and place them into different physical machines (PMs) in the data centers, where R depends upon the server failure probability. We show that VMR is equivalent to the minimum cost flow problem, which can be solved efficiently and optimally. We further reduce the power consumption in the data center by consolidating PMs that store VM replicas and turning off inactive ones. In addition, we design two time-efficient heuristic algorithms to solve VMR. Via extensive simulations and analysis, we compare the VM replication algorithms under different data center scenarios, and show that our consolidation algorithm could further consolidate 50 PMs upon the VM replication algorithm in a data center of 1028 PMs.