Yuan, Hui; (2020) Optical Technologies and Control Methods for Scalable Data Centre Networks. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Yuan_ID_thesis.pdf Download (6MB) | Preview

Abstract

Attributing to the increasing adoption of cloud services, video services and associated machine learning applications, the traffic demand inside data centers is increasing exponentially, which necessitates an innovated networking infrastructure with high scalability and cost-efficiency. As a promising candidate to provide high capacity, low latency, cost-effective and scalable interconnections, optical technologies have been introduced to data center networks (DCNs) for approximately a decade. To further improve the DCN performance to meet the increasing traffic demand by using photonic technologies, two current trends are a)increasing the bandwidth density of the transmission links and b) maximizing IT and network resources utilization through disaggregated topologies and architectures. Therefore, this PhD thesis focuses on introducing and applying advanced and efficient technologies in these two fields to DCNs to improve their performance. On the one hand, at the link level, since the traditional single-mode fiber (SMF) solutions based on wavelength division multiplexing (WDM) over C+L band may fall short in satisfying the capacity, front panel density, power consumption, and cost requirements of high-performance DCNs, a space division multiplexing (SDM) based DCN using homogeneous multi-core fibers (MCFs) is proposed.With the exploited bi-directional model and proposed spectrum allocation algorithms, the proposed DCN shows great benefits over the SMF solution in terms of network capacity and spatial efficiency. In the meanwhile, it is found that the inter-core crosstalk (IC-XT) between the adjacent cores inside the MCF is dynamic rather than static, therefore, the behaviour of the IC-XT is experimentally investigated under different transmission conditions. On the other hand, an optically disaggregated DCN is developed and to ensure the performance of it, different architectures, topologies, resource routing and allocation algorithms are proposed and compared. Compared to the traditional server-based DCN, the resource utilization, scalability and the cost-efficiency are significantly improved.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVJSONXML

Archive Staff Only

View Item