Intelligent fusion identification network state prediction and congestion control system

Abstract

The invention provides an intelligent fusion identification network state prediction and congestion control system. The invention discloses a method and a system for realizing network state predictionand data packet queue congestion control on a programmable data plane by combining a machine learning model method based on P4. The method comprises the following steps: collecting network state feature information in real time through an in-band network telemetry (INT) technology; and achieving network state characteristic value prediction by adopting an 'LSTM-fuzzy clustering' model method combining a long short term memory (LSTM) neural network model and a fuzzy clustering algorithm, and achieving the prediction of the network state characteristic value by adopting a fuzzy clustering algorithm. According to the obtained network state characteristic value, fuzzy clustering is carried out to obtain four network states; a normal state, a congestion early warning state, a continuous congestion state and a congestion alleviation state; corresponding strategies are formulated for different network states, the controller issues corresponding flow tables and formulates switch actions in different network states, and a comprehensive and dynamic queue feedback mechanism is provided to ensure that data packet queuing delay is as small as possible and achieve the effect of congestion control.

Description

technical field [0001] The present invention relates to the field of programmable data plane switch and communication technology, in particular to a network state prediction and congestion control system of Zhirong identification. Background technique [0002] With the rapid development of the Internet in recent years, the number of network users has increased day by day, and the network scale has continued to expand. Especially now that various new network business forms have appeared in large numbers, the problem of network congestion has intensified. In the current network research, how to avoid network congestion, improve network communication performance, optimize the configuration of network resources, and ensure network reliability, real-time performance, stability, robustness and other performances has always been an important issue in the academic and industrial circles. Hot spots of concern. The current congestion control mechanism is composed of end-to-end transm...

AI Technical Summary

Problems solved by technology

[0006] On the one hand, this technology needs to issue statistical policies to the traffic acquisition module of the forwarding layer to obtain traffic during use, which brings a large bandwidth and time overhead; on the other hand, it only predicts and analyzes whether there is a disturbance in the network and Whether it is running normally, to achieve congestion control and load balancing, lack of analysis of the relationship between prediction feature results and their impact on network status, and the contingency of various factors cannot be avoided

Images