Energy-Efficient Dynamic Training and Inference for GNN-Based Network Modeling

TL;DR

The paper tackles the challenge of energy-efficient, context-aware dynamic training and inference of GNNs for network modeling in a mobile-edge-cloud setup. It introduces QAG, a low-complexity framework that uses a tripartite graph to jointly optimize application requirements, GNN configurations, and compute-node assignments through a Quantum Approximate Optimization (QAOA) approach. By pruning infeasible edges and solving a max-cut on the complement graph, QAG finds feasible, energy-saving deployments that meet latency and loss constraints, closely approximating optimal solutions and outperforming a state-of-the-art baseline. The results demonstrate substantial energy reductions (often exceeding 50%) and churn-rate improvements (up to ~80%), validating the practicality of quantum-inspired orchestration for large-scale, heterogeneous network modeling tasks.

Abstract

Efficient network modeling is essential for resource optimization and network planning in next-generation large-scale complex networks. Traditional approaches, such as queuing theory-based modeling and packet-based simulators, can be inefficient due to the assumption made and the computational expense, respectively. To address these challenges, we propose an innovative energy-efficient dynamic orchestration of Graph Neural Networks (GNN) based model training and inference framework for context-aware network modeling and predictions. We have developed a low-complexity solution framework, QAG, that is a Quantum approximation optimization (QAO) algorithm for Adaptive orchestration of GNN-based network modeling. We leverage the tripartite graph model to represent a multi-application system with many compute nodes. Thereafter, we apply the constrained graph-cutting using QAO to find the feasible energy-efficient configurations of the GNN-based model and deploying them on the available compute nodes to meet the network modeling application requirements. The proposed QAG scheme closely matches the optimum and offers atleast a 50% energy saving while meeting the application requirements with 60% lower churn-rate.

Figures (11)

• Figure 1: Dynamic orchestration of GNN-based model training and inference for network modeling and prediction.
• Figure 2: Application-load-resource tripartite graph model. This small-scale example shows orchestration steps of two applications (blue and green edges) in a system with four configs. (combination of two GNN architectures and two data sources), and two computing nodes (CPU and GPU).
• Figure 3: Edge weight for application ($h\in{\cal{H}}$, $v_i\in{\cal{V}}_1$) to configuration ($\sigma\in\boldsymbol{\sigma}$, $v_j\in{\cal{V}}_2$) (left) and configuration to compute node ($n\in{\cal{N}}$, $v_k\in{\cal{V}}_3$) vertices (right) in the tripartite graph.
• Figure 4: (a) Inference-time/sample on compute nodes; and (b) Loss (MAPE) performance (config.: load-and-infer) for the applications listed in Table \ref{['t:app']}.
• Figure 5: (a) Training time/step on compute nodes; and (b) Loss (MAPE) (config.: 20 epochs, 2000 steps/epoch) for applications listed in Table \ref{['t:app']}.