The Dynamic Traffic Monitoring with MAB-Based Sketch Structures for Distributed Network Environments

For efficient network management, effective data traffic monitoring is essential. However, existing monitoring methods often struggle to dynamically adapt to varying traffic characteristics while ensuring high processing speed and query accuracy. This paper proposes a novel sketch structure for data traffic monitoring that integrates a Multi-Armed Bandit (MAB) algorithm. This sketch can dynamically select different subsketches based on the characteristics of the data traffic. The MAB algorithm enables the sketch to learn and adapt to changing traffic patterns in real time, optimizing the selection of the most suitable sub-sketch for different traffic conditions. The proposed sketch enhances measurement accuracy through dynamic adaptation and supports flexible extension for evolving traffic needs. Building on this design, the integration of the MAB algorithm offers a promising solution for efficient and accurate data traffic monitoring in modern networks. To evaluate its effectiveness, extensive experiments were conducted on three distinct datasets. Results show that MAB Sketch significantly improves accuracy, outperforming existing techniques by a factor of 3.33.