Using Distributed Reinforcement Learning for Resource Orchestration in a Network Slicing Scenario

TL;DR

This work tackles dynamic resource orchestration for network slicing in 5G by employing distributed reinforcement learning.A multi-agent architecture deploys Advantage Actor-Critic (A2C) to select continuous actions for bandwidth, compute, and memory resources, coordinated by a central training manager, enabling online adaptation.System performance is captured through slice-level utilities that reflect throughput and delay satisfaction under the assigned resources, with an overall utility aggregating elastic and strict SLA slices across flows.Experiments across Dumbbell, Triangle, and Pyramid topologies show the proposed distributed DRL approach outperforming static and empirical baselines, and transfer learning enabling faster adaptation to new topologies; a GARR-based deployment demonstrates scalability.

Abstract

The Network Slicing (NS) paradigm enables the partition of physical and virtual resources among multiple logical networks, possibly managed by different tenants. In such a scenario, network resources need to be dynamically allocated according to the slices' requirements. In this paper, we attack the above problem by exploiting a Deep Reinforcement Learning approach. Our framework is based on a distributed architecture, where multiple agents cooperate towards a common goal. The agents' training is carried out following the Advantage Actor Critic algorithm, which allows to handle continuous action spaces. By means of extensive simulations, we show that our approach yields better performance than both a static allocation of system resources and an efficient empirical strategy. At the same time, the proposed system ensures high adaptability to different scenarios without the need for additional training.

Figures (11)

• Figure 1: Resource performance function.
• Figure 2: Learning Architecture.
• Figure 3: Link resource allocation.
• Figure 4: Network topologies.
• Figure 5: Training phase.