Predicting Bandwidth Utilization on Network Links Using Machine Learning

Maxime Labonne, Charalampos Chatzinakis, Alexis Olivereau

Abstract

Predicting the bandwidth utilization on network links can be extremely useful for detecting congestion in order to correct them before they occur. In this paper, we present a solution to predict the bandwidth utilization between different network links with a very high accuracy. A simulated network is created to collect data related to the performance of the network links on every interface. These data are processed and expanded with feature engineering in order to create a training set. We evaluate and compare three types of machine learning algorithms, namely ARIMA (AutoRegressive Integrated Moving Average), MLP (Multi Layer Perceptron) and LSTM (Long Short-Term Memory), in order to predict the future bandwidth consumption. The LSTM outperforms ARIMA and MLP with very accurate predictions, rarely exceeding a 3% error (40% for ARIMA and 20% for the MLP). We then show that the proposed solution can be used in real time with a reaction managed by a Software-Defined Networking (SDN) platform.

Introduction

Modern networks rely on reactive mechanisms to handle congestion. Whether it is allocating more bandwidth, prioritizing traffic via Quality of Service (QoS) rules, or load balancing across multiple paths, these actions are typically taken after a bottleneck is detected. By the time these systems react, users are already experiencing packet loss, buffering, and increased latency.

This paper addresses the challenge of proactive network management. By accurately predicting future bandwidth utilization, network administrators or automated Software-Defined Networking (SDN) controllers can reroute traffic or apply rate limits before a link reaches capacity. We demonstrate a machine learning approach using Long Short-Term Memory (LSTM) networks to predict utilization 15 seconds into the future with high precision.

Methods and Materials

The research utilized a simulated network environment built with VirtualBox, generating randomized TCP/UDP traffic via iPerf3. Network state data was captured using Netmate for flow statistics and Dstat for system resource metrics.

Feature Engineering

Because raw flow data is granular, the authors aggregated Netmate statistics using functions such as min, max, mean, sum, and count. This process transformed flow-level data into 86 link-level features, which were then combined with 29 system features from Dstat for a total of 115 features.

Feature engineering process aggregating Netmate and Dstat metrics into 115 total features.

Modeling and SDN Integration

The forecasting task was framed as a supervised time-series problem. Three models were evaluated: ARIMA (a single-feature baseline), MLP (a feedforward neural network), and LSTM (a recurrent neural network). The best-performing model was integrated into a real-time SDN setup using Open vSwitch, where the controller proactively installed flow rules based on 15-second-ahead predictions.

Results and Discussion

The LSTM model significantly outperformed both ARIMA and MLP. The inherent ability of LSTMs to capture complex temporal dependencies and bursty patterns proved crucial for accurate network traffic forecasting. In contrast, ARIMA's reliance on historical bandwidth alone was insufficient for usable results.

Comparison of predicted vs. real bandwidth usage ratio for the LSTM model, demonstrating high accuracy.

Key performance metrics include:

  • LSTM Accuracy: The model achieved an average Root Mean Squared Error (RMSE) of 0.0122 and a Mean Absolute Error (MAE) of 0.0042.
  • Error Margins: LSTM prediction errors rarely exceeded 3% (equivalent to +/- 300 Kbps on 100 Mbps links). MLP and ARIMA errors were significantly higher, often exceeding 20% and 40%, respectively.
  • Proactive Control: In the SDN experiment, the controller successfully used these predictions to trigger load-balancing before links reached 100% capacity, effectively smoothing network performance.

Conclusion

This study proves that ML-driven forecasting is a viable tool for proactive network management. While the current 115-feature model is highly accurate, future work will focus on dimensionality reduction to minimize computational and telemetry overhead. Additionally, the authors plan to validate the system on real hardware with diverse application traffic and expand predictions to include system-level resource exhaustion (CPU and RAM).

Summary updated

Abstract

Predicting the bandwidth utilization on network links can be extremely useful for detecting congestion in order to correct them before they occur. In this paper, we present a solution to predict the bandwidth utilization between different network links with a very high accuracy. A simulated network is created to collect data related to the performance of the network links on every interface. These data are processed and expanded with feature engineering in order to create a training set. We evaluate and compare three types of machine learning algorithms, namely ARIMA (AutoRegressive Integrated Moving Average), MLP (Multi Layer Perceptron) and LSTM (Long Short-Term Memory), in order to predict the future bandwidth consumption. The LSTM outperforms ARIMA and MLP with very accurate predictions, rarely exceeding a 3\% error (40\% for ARIMA and 20\% for the MLP). We then show that the proposed solution can be used in real time with a reaction managed by a Software-Defined Networking (SDN) platform.
BibTeX
@inproceedings{labonne2020predicting,
  author = {Maxime Labonne and Charalampos Chatzinakis and Alexis Olivereau},
  title = {{Predicting Bandwidth Utilization on Network Links Using Machine Learning}},
  booktitle = {EuCNC},
  year = {2020},
  url = {https://openreview.net/forum?id=01dBN9zCMJ},
}
Figure