Graph Neural Network for In-Network Placement of Real-Time Metaverse Tasks in Next-Generation Networks
Delivering realistic, real-time virtual experiences in the Metaverse demands computationally intensive rendering tasks with strict latency constraints. These challenges are further intensified by dynamic behavior and heterogeneity of network resources. While traditional machine learning (ML) methods...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11050417/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Delivering realistic, real-time virtual experiences in the Metaverse demands computationally intensive rendering tasks with strict latency constraints. These challenges are further intensified by dynamic behavior and heterogeneity of network resources. While traditional machine learning (ML) methods have been applied to task placement, they often fail to accurately capture the complex relationships between tasks and computing nodes in such dynamic environments. To overcome these limitations, this study introduces a graph neural network (GNN)- based adaptive framework for in-network placement of delay-sensitive tasks in the Metaverse, leveraging the computing-in-the-network (COIN) paradigm. The task placement challenge is modeled as an integer linear programming (ILP) problem aimed at minimizing overall system costs while adhering to stringent delay constraints. Although optimal solutions can be derived using standard optimization solvers, their high computational overhead makes them unsuitable for real-time deployment. To address this, the optimal ILP solutions train a GNN offline, enabling fast, efficient placement decisions at runtime. The proposed GNN framework captures connected and isolated node interactions, facilitating dynamic adaptation to network conditions. It achieves a placement accuracy of 92.44%, surpassing multilayer perceptron 76.25% and decision trees 70.16%. Simulation results further confirm that the proposed model substantially outperforms conventional ML approaches, while task partitioning across multiple COIN nodes enhances network resource utilization. |
|---|---|
| ISSN: | 2169-3536 |