Large distributed virtual infrastructure partitioning and provisioning across providers

Cloud environments provide elastic capacity and flexible pay-as-you-go business model, and can significantly reduce the operational cost for resource-intensive applications like big data, deep learning and the Internet of Things (IoT). The virtual infrastructure, including networked virtual machines, for large scale applications, is often distributed, and involve resources not only from data centers, but also distributed Fog and Edge nodes. It is not always feasible to use resources all from a single provider. The provisioning of a complex virtual infrastructure requires dynamic partitioning of the topology and seamless configuration of the network, to meet the latency constraints of geolocation devices, to optimize the price and Service Level Agreements of the application. In many cases, the partition solution of a virtual infrastructure cannot be done in the design phase and has to consider the actual resource availability of providers at the provisioning stage. This paper presents a virtual infrastructure partitioning and provisioning solution with consideration of the QoS and resource constraints from the application.