Utilizing TabNet Deep Learning for Elephant Flow Detection by Analyzing Information in First Packet Headers.

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Efficient network traffic management requires rapid detection of significant data streams.
• Identifying large-scale data flows (elephant flows) is critical for optimizing network performance.
• Traditional methods may struggle with the scale and speed of modern network traffic.

Purpose of the Study:

• To investigate the effectiveness of the TabNet deep learning architecture for identifying elephant flows.
• To analyze the 5-tuple fields of initial packet headers for flow characteristics.
• To assess the potential for reducing flow table entries in network devices.

Main Methods:

• Utilized the TabNet deep learning model for flow analysis.
• Analyzed 5-tuple information from the initial packet header.
• Trained and tested the model on a campus network dataset.

Main Results:

• The TabNet model accurately identifies elephant flows at the beginning of their transmission.
• The number of required flow table entries was reduced by up to 20 times.
• Effective management of 80% of network traffic was achieved using individual flow entries.

Conclusions:

• TabNet offers a robust and accurate solution for real-time elephant flow detection.
• This approach significantly enhances network traffic management efficiency and scalability.
• The model's performance on a campus network dataset suggests broad applicability across diverse network environments.