Abstract:
The fine-grained functional split is an effective way to solve the baseband function processing centralization and optical bandwidth saving in radio access networks (RANs...Show MoreMetadata
Abstract:
The fine-grained functional split is an effective way to solve the baseband function processing centralization and optical bandwidth saving in radio access networks (RANs). In this paper, to improve computing resource utilization, we investigate how to realize the fine-grained function placement and routing of 5G RAN slice with function reuse scheme in elastic optical networks (EONs). We first formulate a mixed integer linear programming (MILP) model to solve the problem exactly. The main optimization goal in the MILP model is to jointly minimize the average cost of computing, bandwidth resources and end-to-end latency. Then, a heuristic-assisted deep reinforcement learning (HA-DRL) algorithm is proposed to obtain a near-optimal solution. In particular, the longest common subsequence-based path policy is utilized in the DRL to reduce the size of the action space and accelerate the training process. Finally, we evaluate the proposed MILP model and HA-DRL algorithm via extensive simulation. The results show that the proposed MILP model and HA-DRL algorithm outperform the benchmarks in terms of average cost, including the number of used processing pools (PPs), maximum frequency slot index (MFSI) on the lightpath and end-to-end latency of each slice request.
Date of Conference: 28 May 2023 - 01 June 2023
Date Added to IEEE Xplore: 23 October 2023
ISBN Information:
Electronic ISSN: 1938-1883