Architecture And Scheduling Mechanism Of Wireless Data Center Networks

Data Center Network is the important infrastructure of cloud computing. The per-formance of current Ethernet-based DCN solutions is severely constrained by the con-gestion caused by unbalanced traffic distributions. Due to the fixed topology, the hotspotwhich generates a high volume of traffic cannot obtain enough available bandwidth andthe consequent high delay would further influence the performance of the whole system.Different from Ethernet, wireless is featured with its flexibility so it is possible to exploitwireless to resolve the requirements of those hotspots. This thesis focuses on utilizingwireless networks to improve the performance of data centers. We design a wireless-enabled DCN architecture and investigated the wireless scheduling problem.We discuss the feasibility of employing wireless to support DCN. AEthernet/wireless-hybrid architecture EWHA is proposed to coordinate the Ethernetinfrastructure and supplementary wireless part. Based on the design, we provide adistributed wireless schedule algorithm to adapt the wireless network to the dynamictraffic distributions.For wireless link configuration and channel allocation, which is the key problemin the wireless scheduling mechanism, we formulate the problem based on Protocol-Interference model, which avoid conflicting parallel wireless transmissions, and take theimpact of radios, channels and interference into consideration. Two different approachesare proposed to address problem. The MMS approach aims to minimize the traffic ofhotspotswhiletheMUSapproachmaximizestheutilityofthewirelessnetwork. Agreedyalgorithm and a Maximum-Weighted-Matching-based algorithm are designed for the twoapproaches, respectively. Simulation results shows that our solutions can considerablyimprove the performance of DCN with the help of wireless transmissions.Based on the scheme for Protocol-Interference model, we further extend it to a morepractical Physical-Interference model, which considers different interference strength.We upgrade the MaxUtility approach by involving the signal-to-interference-noise-radioand the adaptive wireless transmission rate. An optimized genetic algorithm GAS, whichcan handle the evolving traffic efficiently, and a relaxing-rounding-based approximationalgorithm RRS, whose performance is higher than the1/2of the optimal solution, areprovided to solve the complicated problem. An extensive simulation-based evaluation demonstrated that both the algorithms can effectively tackle the wireless scheduling prob-lem as well as improve the performance of the whole DCN in terms of throughput and soon.