All You Need is Transformer: RTT Prediction for TCP based on Deep Learning Approach
Runyi Li, Xuan Zhang · 2021
In the Transmission Control Protocol (TCP) connection and transmission process, Round-Trip Time (RTT) is an important indicator, which reflects the speed and reliability of the connection. RTT is equal to the time from start to end of a network transmission between clients and servers, which is vital to study network stability and performance. There are many methods deep-learning-based methods to predict such time series. Among them, the method based on Transformers is the current state-of-the-art. This article (1) proposes a multi-step RTT sequence prediction model and used the teacher forcing mechanism. The model is based on Transformers, which can predict the RTT of a sequence over a period, and achieves good accuracy on the dataset captured by ourselves; (2) uses a new deep learning method, which was proved the effectiveness of Transformers on time series tasks and creates a new paradigm for computer network research.