Network Information Flow Delay Prediction Based On InbandNetwork Telemetry and Deep Learning

Network delay is a basic measure of computer network QoS. Accurate network delay prediction can provide a very necessary technical support for the formulation of real-time network application strategy. Unfortunately, the traditional network delay prediction technology can not meet this requirement. This paper focuses on the study of fine-grained time delay prediction for network information flow, which is typical spatio-temporal data. Firstly, the fine-grained parameters of network information flow are obtained and transformed by using the in-band measurement technology. Secondly, Combining with the spatio-temporal characteristics of network information flow, this paper proposes a multi-time scale fusion network delay prediction model, namely Nefopame, which is based on improved LSTM and GCN. This model accurately captures the spatio-temporal correlation of network information flow, and analyzes and predicts the network queuing delay. Experiments on real network information flow data sets show that our Nefopame achieves more accurate delay prediction results than other benchmark programs.