G Network Slicing: Dynamic Resource Management For Ultra-Reliable Communication

Abstract

Next-generation G networks will enable mission-critical services that will be supported by abstract-reliable communication and at the same time be able to support heterogeneous traffic that will have varied quality-of-service needs. Network slicing has been one of the enablers that enable logical separation of shared physical infrastructure into service specific virtual networks that are isolated. The paper introduces a dynamic resource management system of G network slicing in an attempt to provide ultra-reliable, low-latency, and adaptive communication in time-varying traffic conditions. The framework proposed combines the slice level resource abstraction with smart utilization of bandwidth, transmission power and edge cloud computing resources. The recurrent learning models, such as LSTM-TDP and GRU-NetSlice, are used to predict traffic and demand dynamics, which allow a proactive slice provisioning. To resolve real-time uncertainty, deep reinforcement learning agents, i.e., DQN-Slice and PPO-SliceManager are used to optimize the policies of slice that maximize the reliability, spectral efficiency and resource utilization simultaneously. Reliability-conscious optimization models are added to provide probabilistic latency and packet delivery constraints that are needed in ultra-reliable low-latency communication services. In addition, redundancy-based and fast slice reconfiguration-based fault tolerance mechanisms are presented to improve service availability in the event of failures. The evaluation by simulation shows that the proposed intelligent slicing framework is much more reliable, less end-to-end latency, and more adaptable to network dynamism than the existing, and the heuristic allocation schemes.