Neural-based TCP performance modelling

Web users are expecting shorter response time when they are using Internet. However, Internet traffic is continued growth by adding more services and functions, the traffic will cause congestion problems and delays on Internet. In this paper, we introduce the TCP protocol theory, the different between TCP long-lived and short-lived connection. And the neural network structure, algorithm and the parameters. Also described the current state of TCP short-lived connection traffic analysis and performance modelling. We explored and compared existing models for TCP long-lived and short-lived connection data transfer latency, the advantages and disadvantages are discussed. This paper proposes to use mathematical model and neural network to predict TCP transfer latency for short-lived TCP connection for non-packet loss and packet loss situations, the results are compared by using the relative error and the overall comparison