Performance Analysis of Parallel Transmission and Multipath Routing in High-Speed Network Systems

Abstract

The unprecedented growth in Internet traffic is driving the steep upscaling of network capacity, which even in optical networks is expected to reach the so-called Shannon limit. The capacity challenge has thus led to innovations in all areas of networking. One of such innovations is the parallelization of end-systems combined with space division multiplexing in the network. As a result, parallel transmission combined with multipath routing is identified as a key solution to address the imminent capacity crunch and harvest the power of end-system hardware capabilities. The newly discovered benefits of optical Orthogonal Frequency Division Multiplexing (OFDM) networks are also driving the need to jointly design and analyze network system parallelism and multipath routing. This thesis sets the goal to uniquely tackle the challenges associated with network capacity upscaling by unifying the research on multipath routing and parallel network system design. It starts by modeling multipath routing problems for data-intensive applications and thereafter identifies challenges and benefits of the same, while proposing novel and practically relevant solutions. Furthermore, the thesis presents pioneering studies in high-speed Ethernet parallel transmission in combination with various optical network technologies, from conventional WDM networks to advanced optical OFDM networks. To this end, this thesis presents the first attempt to validate that high-speed Ethernet standard can benefit from the inherent parallelism of optical OFDM networks without resource penalty, such as spectrum fragmentation. Finally, this thesis addresses the critical issue of buffer dimensioning in high-speed parallel network systems, and outlines efficient solutions to address the same. Proposed novel solutions in combination with linear network coding are shown to hold the key to an efficient high-speed Ethernet system design as they can significantly lower the buffer requirement at the receiver.