Wireless 60 GHz Rack to Rack Communication in a Data Center Environment.

摘要

Data centers play an increasingly important role in processing the large amount of information generated in today's society. An enormous amount of growth in the computational demands of data center applications has stimulated the creation of warehouse scale data centers, holding servers that number in the thousands. As the number of servers within a data center grows, the interconnecting infrastructure becomes of paramount importance. Present day interconnects are formed using either copper wire in a twisted pair configuration or through the use of fiber optic cables. One of the main concerns with the scalability of a data center's interconnecting network is the power consumption. Large power hungry switches at the aggregation and core levels make up a significant portion of a data centers power portfolio and cannot be overlooked. Furthermore, large bundles of wires both reduce the air flow within data centers and are costly to replace and maintain. This cabling complexity problem limits cooling effectiveness and exacerbates the power consumption challenges.;Recent advancements in the unlicensed 60 GHz spectrum have given rise to transceivers that can support high bandwidth links, comparable to wired links found in most data centers. These wireless links also exhibit promising characteristics such as spatial reusability which make them suitable within a data center environment. By taking advantage of emerging 60 GHz wireless technologies, data centers can utilize these high speed wireless links to satisfy bandwidth demands while simultaneously reducing their power consumption and cabling requirements.;This thesis evaluates the benefits in terms of energy-efficiency of using 60 GHz wireless links to replace wire line links within a data center by modeling a completely wireless data center. The physical layer design and associated MAC layer will be investigated to support this wireless centric design. The proposed wireless architecture will be compared against traditional hierarchical data center architectures and evaluated based upon several performance metrics such as throughput, latency, and overall energy efficiency.