Network coding for computer networking

Conventional communication networks route data packets in a store-and-forward mode. A router buffers received packets and forwards them intact towards their intended destination. Network Coding (NC), however, generalises this method by allowing the router to perform algebraic operations on the packets before forwarding them. The purpose of NC is to improve the network performance to achieve its maximum capacity also known as max-flow min-cut bound. NC has become very well established in the field of information theory, however, practical implementations in real-world networks is yet to be explored. In this thesis, new implementations of NC are brought forward. The effect of NC on flow error control protocols and queuing over computer networks is investigated by establishing and designing a mathematical and simulation framework. One goal of such investigation is to understand how NC technique can reduce the number of packets required to acknowledge the reception of those sent over the network while error-control schemes are employed. Another goal is to control the network queuing stability by reducing the number of packets required to convey a set of information. A custom-built simulator based on SimEvents® has been developed in order to model several scenarios within this approach. The work in this thesis is divided into two key parts. The objective of the first part is to study the performance of communication networks employing error control protocols when NC is adopted. In particular, two main Automatic Repeat reQuest (ARQ) schemes are invoked, namely the Stop-and-Wait (SW) and Selective Repeat (SR) ARQ. Results show that in unicast point-to point communication, the proposed NC scheme offers an increase in the throughput over traditional SW ARQ between 2.5% and 50.5% at each link, with negligible decoding delay. Additionally, in a Butterfly network, SR ARQ employing NC achieves a throughput gain between 22% and 44% over traditional SR ARQ when the number of incoming links to the intermediate node varies between 2 and 5. Moreover, in an extended Butterfly network, NC offered a throughput increase of up to 48% under an error-free scenario and 50% in the presence of errors. Despite the extensive research on synchronous NC performance in various fields, little has been said about its queuing behaviour. One assumption is that packets are served following a Poisson distribution. The packets from different streams are coded prior to being served and then exit through only one stream. This study determines the arrival distribution that coded packets follow at the serving node. In general this leads to study general queuing systems of type G/M/1. Hence, the objective of the second part of this study is twofold. The study aims to determine the distribution of the coded packets and estimate the waiting time faced by coded packets before their complete serving process. Results show that NC brings a new solution for queuing stability as evidenced by the small waiting time the coded packets spend in the intermediate node queue before serving. This work is further enhanced by studying the server utilization in traditional routing and NC scenarios. NC-based M/M/1 with finite capacity K is also analysed to investigate packet loss probability for both scenarios. Based on the results achieved, the utilization of NC in error-prone and long propagation delay networks is recommended. Additionally, since the work provides an insightful prediction of particular networks queuing behaviour, employing synchronous NC can bring a solution for systems’ stability with packet-controlled sources and limited input buffers