Optimising the design of passive optical networks-based data centres

Abstract

The recent growth in cloud-based applications has motivated researchers to focus on improving the scalability, power efficiency, and cost-effectiveness of data centre architectures. State-of-the-art data centres consists of numerous access and aggregation switches which can be costly and can lead to inefficiencies such as unbalanced traffic and high-power consumption. Passive Optical Network (PON) technology, known for its high performance in access networks, can offer energy-efficient, cost-effective, and scalable solutions for modern and future data centres. This thesis aims to propose and enhance a PON-based data centre architecture to enable multi-path routing, load balancing and scalability, enhance resilience and energy efficiency, and reduce latency. The proposed design is based on a two-tier cascaded Arrayed Waveguide Grating Routers (AWGRs) fabric. We develop a Mixed Integer Linear Programming (MILP) model to optimise the wavelength assignment and multipath routing in the proposed architecture. Additionally, we investigate the resilience of the proposed architecture by evaluating its power consumption and delay under several failure scenarios. We also optimise virtual Machine (VM) placement in the proposed architecture to minimise power consumption. A MILP model and a heuristic are developed for VM placement in the proposed architecture and the results show significant power consumption reductions, up to 66% compared to VM placement in the state-of-the-art architecture, the spine and leaf architecture. Furthermore, we consider the use of WDM/TDM multiple access technique with multipath routing in the proposed architecture. A MILP model is developed to jointly optimise time slots allocation and routing and wavelength assignment.