Disaggregated Servers for Future Energy Efficient Data Centres

The popularity of the Internet and the demand for 24/7 services uptime is driving system performance and reliability requirements to levels that today’s data centres can no longer support. This thesis examines the traditional monolithic conventional server (CS) design and compares it to a new design paradigm known as disaggregated server (DS). The DS design arranges data centres resources in physical pools such as processing, memory and IO module pools; rather than packing each subset in a single server. In this work, we study energy efficient resource provisioning and virtual machine (VM) allocation in the DS based data centres compared to CS based data centres. First, we developed a mixed integer linear programming (MILP) model to optimise VM allocation for DS based data centre. Our results indicate that considering pooled resources yields up to 62% total saving in power consumption compared to the CS approach. Due to the MILP high computational complexity, we developed an energy efficient, fast and scalable resource provisioning heuristic (EERP-DS), based on the MILP insights, with comparable power efficiency to the MILP. Second, we extended the resources provisioning and VM allocation MILP to include the data centre communication fabric power consumption. The results show that the inclusion of the communication fabric still yields considerable power savings compared to the CS approach, up to 48% power saving. Third, we developed an energy efficient resource provisioning for DS with communication fabric heuristic (EERP-DSCF). EERP-DSCF achieved comparable results to the second MILP and with it we can extend the number of served VMs where the MILP scalability for big number of VMs is challenging. Finally, we present our new design for the photonic DS based data centre architecture supplemented with a complete description of the architecture components, communication patterns and some recommendations for the design implementation challenges.