Massive-scale deployments in cloud

| openaire: EC/H2020/644866/EU//SSICLOPSThe current trends in technology are shifting towards cloudification of services and massive device connectivity. Meanwhile, cloud technology is undergoing a transition from conventional full-fledged virtual machines towards lightweight container-based solutions. This results in a heavy reduction of the virtualization overhead allowing to multiply the number of running instances. There is a need for highly scalable and efficient cloud infrastructures. Our study focuses on OpenStack, as an Infrastructure-as-a-Service provider, and in particular the case of large service deployments on a single node. We conduct a comprehensive analysis of the different network building blocks used by OpenStack. This reveals important scalability limitations which negatively impact the network performance of the node and cause an unreasonably high CPU usage. We identify some of the root causes for these scaling constraints and propose a number of alternative designs. Throughout both analytical and empirical methodologies, we demonstrate the benefits of our solutions in several aspects such as scalability, efficiency and performance.Peer reviewe