Automated controllers for bandwidth allocation in network virtualization

The concept of network virtualization was introduced to facilitate flexible service deployment for the future Internet. This recent technology provides a powerful tool to run multiple logical networks on the same physical substrate defined as virtual networks (VNs). Each physical link is split into virtual links and each VN receives a fraction of the available capacity. Bandwidth allocation for multiple VMs aims at sharing the physical links among multiple VNs. It is a critical challenge for both service providers (SPs) and infrastructure providers (InPs). This allocation should take into account the quality of service (QoS) requirements of the flows that are crossing each VN. In this paper, we consider a virtualized network environment where the SP deploys multiple VNs with different links' capacity demands and QoS requirements. Each VN competes with other VNs to receive fractions of physical links managed by multiple InPs. We present a two-layer controller system that adapts to the dynamic change of the workload of each VN. The system uses a prediction-based approach in order to find the optimal request for each VN. The request depends on the estimation of the relationship between the VN performance in terms of packet delays and the actual and past allocations. Then, due to the capacity constraint of the physical link, the system adjusts the offered bandwidth for each of them. Our model offers flexible distributed autonomous control of the bandwidth allocation to maintain the offered QoS to each VN at the desired level in response to the dynamics of the workload. Our mechanism provides an optimum allocation of the physical links by distributing the bandwidth periodically. It also offers the possibility of adjusting the VNs' parameters to take into account the current network behaviour to avoid bottleneck virtual links.