Adaptive-INT: Fast Network Anomaly Perception Based on Multidimensional Attribute Measurement

Network measurement technology has become an indispensable tool in modern networks. By monitoring and managing network states, it ensures the stability and efficiency of network operations. In-band Network Telemetry (INT), based on programmable data planes, enables fine-grained, real-time monitoring of the data plane. It is regarded as a promising measurement technology. However, existing INT technologies struggle to meet diverse monitoring requirements. They face issues such as delayed information and missing critical perceptual data. To address these challenges, we propose a novel network telemetry scheme based on multi-dimensional attribute metrics, named Adaptive-INT. Specifically, Adaptive-INT first introduces a lightweight INT packet structure to reduce the overhead associated with collecting network state information. Next, it uses a path planning algorithm to generate detection paths. This approach ensures information timeliness while minimizing path coverage redundancy. Finally, a dynamic node stability control algorithm is applied to manage the acquisition actions for different types of information from network nodes. Experimental results demonstrate that Adaptive-INT not only effectively reduces network resource consumption, but also provides rapid sensing capabilities, significantly improving the flexibility and responsiveness of network telemetry.