Cost-Aware Neural Adaptive Scaling for vRAN Resource Allocation

초록

Although virtualized radio access networks (vRANs) offer flexibility and scalability, current scaling methods in vRANs tend to overlook the costs associated with energy consumption and service interruptions during the reconfiguration of containerized network functions (cNFs), leading to inefficient resource utilization. Moreover, traditional vertical and horizontal scaling approaches exacerbate these issues by requiring cNFs to be stopped and restarted or by deploying cNFs with fixed resource sizes. To address these challenges, we propose a cost-aware neural adaptive scaling (CNAS) framework, which adjusts cNF resource allocation dynamically, minimizing service disruptions and avoiding overprovisioning. The cost model incorporates energy consumption during the activation, allocation, and termination of cNFs and servers, as well as the operational cost of maintaining active cNFs and servers. We then formulate the integer linear programming (ILP) problem to minimize total costs. Due to the NP-hard complexity of this problem, two heuristic algorithms are applied: one utilizes dynamic programming to establish the cost-aware resource allocation, while the other uses a greedy approach to handle cNFs and servers following the determined resource allocation. Trace-driven simulation results demonstrate that CNAS can reduce the scaling costs by 53.4% and the total costs by 21.3% compared to the state-of-the-art methods.

키워드