Bandwidth-Aware Resource Collaborative Scheduling Method in Fusion Networks

Introduction

With the rapid development of information and communication technologies, future 6G networks aim to provide rich immersive experiences by enabling ubiquitous connectivity through collaboration and convergence. Network slicing (NS) technology, which logically separates physical resources to support customized services and isolation over shared infrastructure, has emerged as an ideal solution for deploying 5G private networks and is considered a key enabler for 6G. Despite various resource allocation algorithms, the challenge of adapting to the unique characteristics of 5G and 6G networks while focusing on “collaboration” remains a critical research gap.

Existing studies have explored network slicing in relation to the three 5G application scenarios defined by 3GPP: enhanced Mobile Broadband (eMBB), massive Machine-Type Communications (mMTC), and Ultra-Reliable Low-Latency Communications (URLLC). Some approaches employ deep reinforcement learning for automated resource adjustment, while others leverage Software-Defined Networking (SDN) and Network Functions Virtualization (NFV) to maximize slice efficiency. However, these methods often lack explicit consideration of resource collaboration.

In fusion networks, such as satellite-terrestrial and telecom-cloud integrations, resource scheduling remains largely static, with limited emphasis on collaborative optimization. This paper addresses the growing demand for bandwidth-intensive applications like high-definition video streaming, augmented/virtual reality (AR/VR), and telemedicine by proposing a Bandwidth-Aware SFC Mapping Algorithm (BASA). The algorithm dynamically coordinates node and link mapping strategies to enhance network performance and Quality of Service (QoS).

System Model and Problem Formulation

System Model

Network slicing allows the creation of multiple logical networks over shared infrastructure, tailored to specific Service-Level Agreement (SLA) requirements. This study focuses on the core network (CN) domain, where the physical substrate network (SN) is modeled as an undirected weighted graph. Similarly, Service Function Chains (SFCs) are represented as virtual networks consisting of interconnected Virtual Network Functions (VNFs).

Users are randomly distributed within base station (BS) coverage areas, and their traffic traverses predefined SFCs. Each SFC must successfully map all required VNFs to physical nodes and establish data flow paths for a user to access the network. Key notations include:
• Physical Network (SN): Comprising server nodes, links, computational capacity, and bandwidth.

• Virtual Network (SFC): Consisting of VNFs, virtual links, computational demands, and bandwidth requirements.

Problem Description

SLA Decomposition

End-to-end SLAs are decomposed into domain-specific requirements. Typical SLA parameters include latency, bandwidth, reliability, and throughput. For bandwidth, the CN must ensure that the actual allocated bandwidth meets or exceeds the minimum threshold required for QoS.

Bandwidth-Aware Resource Collaborative Scheduling

The SFC mapping process involves two phases:

Node Mapping: Virtual nodes are mapped to physical nodes based on computational capacity and bandwidth availability.
Link Mapping: Virtual links are embedded onto physical paths, prioritizing higher bandwidth links. If multiple paths have equal bandwidth, the shortest path is selected.

Key performance metrics include:
• Quality of Experience (QoE): The ratio of successfully served users to total users.

• Bandwidth Utilization: The proportion of actual bandwidth usage to total available bandwidth.

• Bandwidth Fluctuation: Measures stability in bandwidth usage over time.

• User Satisfaction: Compares actual bandwidth received to the minimum required.

• SFC Mapping Success Rate: The ratio of successfully deployed SFCs to total requests.

Bandwidth-Aware Resource Collaborative Scheduling Method

Node Mapping Phase

A pointer network combined with Long Short-Term Memory (LSTM) and attention mechanisms is employed to determine the optimal node mapping sequence. The pointer network processes physical node features and generates a probability distribution for mapping virtual nodes. The attention mechanism ensures that computational capacity and bandwidth are considered, minimizing hotspots and improving efficiency.

Link Mapping Phase

After successful node mapping, an enhanced Yen’s algorithm is used for link embedding. The algorithm prioritizes paths with the highest available bandwidth. If multiple paths meet the bandwidth requirement, the shortest path is selected. This approach ensures efficient resource usage while maintaining QoS.

Reinforcement Learning Framework

The overall SFC mapping process is framed as a reinforcement learning problem, where the agent learns to maximize rewards by balancing user access and satisfaction. The reward function combines QoE and user satisfaction metrics, guiding the agent toward optimal mapping strategies.

Algorithm Complexity

The node mapping phase has a time complexity dominated by the pointer network and attention mechanism. The link mapping phase, using the modified Yen’s algorithm, contributes additional complexity. Overall, the algorithm efficiently scales with network size.

Simulation and Performance Analysis

Simulation Setup

Experiments were conducted in a simulated core network environment using a Fat-tree topology. Key parameters included:
• Physical Network: 100 nodes, 600 links, computational capacity, and bandwidth ranges.

• Virtual Network: 2000 VNFs, SFC requests with varying node counts, and bandwidth demands.

Comparative Algorithms

BASA was compared against:

PN-SFC: Uses pointer networks for node mapping and shortest-path algorithms for link embedding.
TVM-DQN: Leverages deep Q-learning with a tidal virtual machine mechanism for dynamic resource adjustment.

Results

SFC Mapping Success Rate: BASA outperformed PN-SFC and TVM-DQN, maintaining high success rates even with increasing user requests. The attention mechanism and bandwidth-aware link embedding contributed to this robustness.
Bandwidth Fluctuation: BASA exhibited higher but stable bandwidth fluctuations due to its preference for high-bandwidth paths. TVM-DQN showed lower fluctuations owing to its adaptive resource management.
Bandwidth Utilization: BASA achieved superior bandwidth utilization, averaging over 75%, compared to TVM-DQN (70%) and PN-SFC (61.7%).

Conclusion

This paper presented a Bandwidth-Aware Resource Collaborative Scheduling Algorithm (BASA) for fusion networks, focusing on dynamic node and link mapping to enhance QoS. By integrating pointer networks, attention mechanisms, and modified Yen’s algorithm, BASA achieved high SFC mapping success rates and bandwidth utilization.

Future work will extend the algorithm to multi-domain scenarios and incorporate additional SLA metrics such as latency and throughput. Addressing these challenges will further improve the adaptability and efficiency of resource scheduling in next-generation networks.

doi.org/10.19734/j.issn.1001-3695.2024.05.0181

Was this helpful?

0 / 0