Novel Hybrid Optical-/Electrical-Switched Networks for Energy-Efficient Operation

Ubiquitous Information and Communication Technology (ICT) devices and services are consuming a significant and growing portion of global power supplies. Increased energy usage causes economic and environmental problems. Communication networks consume around one fourth of the total ICT energy consumption; therefore, energy efficiency of network equipment requires immediate attention. Optical switching technologies have the potential to provide high network capacities in an energy-efficient manner. In this dissertation, we study novel hybrid optical-/electrical-switched network architectures for energy-efficient operation.

For large enterprise access links, we propose a two-wavelength design. In our design, one wavelength is used as part of a lower-rate static circuit for general-purpose IP traffic, while the second wavelength is dynamically configured into a high-rate access-link circuit for large dataset transfers whenever needed. A few provider-router ports are shared among a larger number of customers given that large dataset transfers are relatively infrequent. This leads to potential start-time delays, but results in significant power and cost savings. We compare two solutions for sharing high-speed provider ports, i.e., Immediate-Request (IR) mode solution and Advance-Reservation (AR) mode solution; the latter requires provider-side storage. Simulation results show that the AR-mode solution can achieve performance improvements over the IR-mode solution in terms of blocking probability and average response times. We also provide a differential cost-and-power comparison of the AR-with-storage mode and the IR mode to quantify the extra cost and power consumption introduced by the in-network storage needed with the AR mode.

For Data Center Networks (DCN), we propose an Optical Switch in the Middle (OSM)
hybrid electrical-packet/optical-circuit architecture. OSM features storage with the core Electrical Packet Switch (EPS), AR scheduling enabled by a multilayer SDN controller and Layer-2 multicast ability. Further, we identified Hadoop MapReduce applications as suitable for utilizing high-speed optical circuits despite their high reconfiguration delay, and proposed four Hadoop modifications for Hybrid Networks (HHN) to enable the effective operation of Hadoop in hybrid DCN architectures. Numerical results validate our hypothesis that it is feasible to achieve similar system-level and user-level performance with HHN, while simultaneously achieving power and cost savings with the hybrid network, when compared to original Hadoop on an EPS-only DCN.