Network Assurance in Intent Based Data Center Networking: A Domain Shift Approach
A look at domain shift in virtual machine resource prediction for intent-based data centers, and at which models stay robust when each machine behaves differently.
- Authors
- Steve Lévesque, Xiaoang Zheng, John Violos, Aris Leivadeas (ÉTS)
- Venue
- IEEE NoF 2025, Montreal, Canada
- Year
- 2025
- Pages
- 81–85
- Keywords
- Network Assurance, Intent Based Networking, Machine Learning, Transformers, Few-shot Learning
- Status
- Published
Overview
Intent-based networking relies on network assurance: continuous monitoring and automated adjustments that keep performance reliable. In a data center, virtual machine (VM) resource utilization metrics are key to predicting network behavior. Yet each VM has its own usage patterns, shaped by its applications, its hardware and its users, and these differences can lead to what machine learning calls domain shift.
Approach
My co-authors and I first frame domain shift as a challenge for VM resource utilization prediction. We then test how robust state-of-the-art models remain under this shift: several recurrent neural networks (RNNs), Transformers and Informers, as well as lightweight zero-shot and few-shot approaches, through extensive experiments on a real-world dataset.
Findings
The Gated Recurrent Unit (GRU) model generalized best while keeping a low computational footprint.
Publication details
Presented at the International Conference on Network of the Future (NoF 2025), held from September 30 to October 3, 2025, in Montreal, Canada, and published by IEEE.