A machine learning approach to TCP state monitoring from passive measurements

Abstract

This dissertation is briefly organized as follows.Chapter 1 Introduction : provides a detailed overview of motivation, and the use cases that are relevant to this dissertation. It also describes the scientific methods used in this dissertation.Chapter 2 Summary and Contributions : presents a brief summary of the included papers in this dissertation and their scientific contributions which have been published in journals and international conferences.Chapter 3 Background : sketches some of the basic contextual background information that provides an overview of the background relevant to the reader and challenges that are dealt with in the research.Chapter 4 Related Work : presents a summary of the relevant state-of-the-art related works found in the literature for the three main use cases we presented in Chapter 1.Chapter 5 Conclusions : provides a summary of the research and highlights the main contributions of the dissertation. This chapter also provides considerations of the research and suggestions for promising future research directions.

Recurrent Neural Network-Based Prediction of TCP Transmission States from Passive Measurements

“…Optimality: As it is shown in Tables XIII and XIV, the experimental results show that our LSTM-based prediction model is able to outperform our previous approach using machine learning techniques [10]. Our LSTM-based TCP variant prediction model achieves accuracies of 97.22%, 96.66% and 94.44% on the emulated, realistic and combined scenario settings, outperforming the standard ML-based which yields accuracies of 93.51%, 95% and 91.66% respectively.…”

“…Therefore, what the intermediate monitor sees may not be exactly what the sender or the receiver sees. The set of methodological challenges we identify involved in performing inference of TCP per-connection states related to network congestion from passive measurements are presented more in detail in [10]. In this paper, we advocate that LSTM-based approaches can give a better prediction accuracy of TCP sender connection states from passive measurements collected at an intermediate node by addressing the aforementioned practical challenges.…”

Learn-ing-Based On-AP TCP Performance Enhancement

“…Some research works have considered the regression of congestion windows on intermediate nodes [44][45][46]. As the TCP senders control the data rates based on TCP state machines maintained in TCP stacks, shadow-state machines are implemented on intermediate nodes to follow the dynamics of the real state machines on the sender sides [45,46].…”

“…As the TCP senders control the data rates based on TCP state machines maintained in TCP stacks, shadow-state machines are implemented on intermediate nodes to follow the dynamics of the real state machines on the sender sides [45,46]. Recently, research works in [44] indicate that these state-machine-based methods have performance and compatibility problems. Furthermore, the authors of [44] utilize the number of unacknowledged packets (UNA) to regress the congestion windows on intermediate nodes.…”

Application-Level Packet Loss Rate Measurement Based on Improved L-Rex Model

“…Under the current TCP/IP architecture, packet loss is repaired with retransmission. Accordingly, it is almost a natural idea to take retransmissions as building blocks to construct loss estimation algorithms [14][15][16][17][18][19]. Likewise, L-Rex is also built on the retransmission patterns.…”