A stable matching method for cloud scheduling

Cloud computing has been one of the revolutionary breakthroughs of this decade in the ICT world and its popularity is soaring more than ever. More and more data centers are being deployed in order to accommodate the physical resources needed by cloud systems. As an important side effect the global energy demand of data centers are also on the rise. In the meantime the advancement in virtualization technologies has made migrating virtual machines from one host to another without shutting them down possible. Therefore the optimization of data center operations through the dynamic placement of virtual machines became a reality. This paper formalizes the well-studied cloud scheduling problem in a matching theoretical model in which the virtual machine to physical server mapping is translated into a stable matching problem. We build on an advanced algorithm from the matching theory domain in order to find the most accommodating scheduling arrangement. Hindered by the complexity of the algorithm, we evaluate various heuristics in numerical simulations of cloud environments. After the verification of the selected heuristic algorithm, we present the implementation of the proposed method as a custom compute scheduler for OpenStack.