Optimal Virtual Machine Placement in Large-Scale Cloud Systems

In this work, we focus on the problem of virtual machines (VMs) placement in geographically distributed data centers. We consider communicating VMs assigned to data centers that are connected over an IP-over-WDM network. We aim to plan and optimize the placement of VMs in data centers so as to minimize the IP-traffic within the backbone network. Thus, we propose first, a formulation which can be considered as a variant of the Hub Location problem modeling and we show its extreme difficulty for medium and large size instances. In order to overcome this difficulty, we reformulate the problem by multi-commodity flow, adopt variable aggregating methods and add valid inequalities to strengthen this new formulation. The different experiments that we present show the effectiveness of our last model in terms of running time and computational resources.