Abstract
The Internet has revolutionized the way that people interact with communication services by making available the services anytime and anywhere. However, access to the Internet can be limited particularly in isolated or less-favored areas due to economic or geographical reasons. Lack of the access to the Internet is often referred to as digital divide, an imaginary barrier that prevents the access to the Internet. The quickest way to break this barrier is to deploy satellite networks. However, the Internet cannot operate efficiently over the satellite networks due to several unique characteristics inherent to satellite links, i.e. high capacity, long latency and frequent channel error. These affect the performance of TCP, a networking protocol on which most Internet applications and services are based. As a result, deriving mechanisms to support QoS enabled transport for the ubiquitous Internet is the main motivation of this research. The research begins with the development of TCP Module, i.e. a simulation module of INET simulation framework, by reusing real-world TCP codes of Linux TCP/IP network stack. The significance of TCP Module rests in the ability to emulate precisely the operational behaviors of real-world TCP. In addition, it creates a realistic simulation platform for implementing and testing real-world congestion control protocols as well as for evaluating the performance of different Linux TCP variants for different simulated network environments. Based on the new platform, a novel end-to-end TCV-delta for a high-speed long-distance lossy connection is proposed. TCP-delta is capable of adapting its operational characteristics to several network factors, i.e. link utilization, link congestion, window size and round trip time. Study shows that TCP-delta exhibits remarkable performance in term of RTT-scalability and BER-, robustness in comparison to other Linux TCP variants. Moreover, two novel TCP-aware resource management techniques, i.e. Small Window First Out Last Drop and Predictive Fair Bandwidth on Demand, are proposed in order to resolve overall performance degradation problem caused by slow throughput convergence and unfair resource consumption of multiple TCP connections competing over the same bottleneck link. Both techniques incorporate cross-layer collaboration between the transport layer and the lower layers to optimize the performance. Study shows that these techniques improve TCP performance in term of friendliness and fairness to the level that is not normally achieved by TCP alone.