UC Santa Cruz

This thesis introduces a number of modifications and enhancements to the Transmission Control Protocol (TCP) aimed at addressing some of the known limitations of TCP taking advantage of larger amounts of working memory capacity at the nodes participating in TCP sessions, and network coding. These TCP variants are TCP-EWSC (TCP Enhanced Wireless Santa Cruz), TCP-NEWT (TCP Network-Coding Enabled Window Transformation), TCP-PNC (TCP Predictive Network Coding) and TCP-RTA (TCP Real-time Topology Adaptiveness). TCP-EWSC brings several capabilities to better adapt traditional TCP for hybrid networks including the capability to respond to sporadic temporary wireless signal outages in a resilient manner with proactive spoofing of receiver zero-window. TCP-NEWT introduces a mechanism for network coding to proactively address packet loss without retransmissions while accurately reporting all the observed TCP flow metrics back to the host. TCP-PNC augments TCP-NEWT with real-time prediction of the expected goodput (packets delivered/total packets sent) and proactive compensation for the same. TCP-RTA dynamically detects a topology change and adapts with an appropriate congestion-control strategy to maximize the effective use of the total available bandwidth. TCP-RTA also factors in the relative impact of the degree of change in topology versus that of the underlying transmission medium to identify the best congestion control strategy to ensure optimal usage of the end-to-end network infrastructure. TCP-EWSC performance results show that end-end throughput is sustained close to the prior levels even after the introduction of link-layer errors, when compared to other TCP variants which show a drop of about 8%. Performance results with TCP-PNC with network coding, show an overall increase of end-to-end TCP throughput in the range of 30-35% depending upon the number of wireless-link-layer losses, without changing the underlying "network congestion". Performance results with TCP-RTA indicate a throughput increase of more than 35% in scenarios involving dynamic topology changes during a TCP session.