Slice admission control using overbooking for enhancing provider revenue in 5G Networks

Network Slicing (NS) provides a new business model to the Infrastructure Providers (InP), where multiple tenants request the InP to provide services to its customers. InP decides whether to accept the request or not to maximize the overall profit due to resource limitation. It is generally seen that tenants over-estimate their slices' maximum requirements. In particular, for elastic slices that have flexible resource requirements, we use the concept of overbooking. Here, we accept slices above actual resource availability. We present a Slice Admission architecture where, a Slice Forecasting Agent (SFA) predicts the future resource (CPU, Memory, Bandwidth) usage of currently active elastic slices for next time window. This predicted information is used by an opportunistic overbooking heuristic, where the system allocates the required resources to each slice. After this, we address the admission control problem using a reinforcement learning (RL) approach that decides to accept/reject an incoming slice request. The performance of our proposed work is compared against three other heuristics (Basic, Prediction, Prediction-RL) that do not use overbooking. Data traces from the Materna data center network were used for prediction. The results show that the proposed work significantly outperforms the other mechanisms in terms of revenue gain and resource utilization.