Optimization of geo-distributed datacenters

Abstract

The increasing interest and investment in data centers clashes with the core goals of the energy transition movement. This is because data centers are enormous electricity consumers, carbon emissions producers and require an abundance of water for cooling. The core goals of the energy transition are decarbonization, sustainability energy efficiency and electric grid security. It is easy to see how the data centers and the energy transition movement clash. Thus, exploring a scenario where the two co-exist is warranted.

This thesis utilized Naug’s et. al. developed tool “SustainCluster”, which aims to provide a benchmark for geo-distributed data center simulation. However, this thesis reviews their background information to gain an understanding of SustainCluster’s functionality. Naug et. al. utilized several studies to model the power consumption of a data center and developed a soft actor-critic algorithm that seeks a policy that adheres to the given objectives. This thesis applied some of the goals of the energy transitions as objectives to optimize and produced results via the benchmark.

It was observed that there exists a scenario where the energy transition and data center can co-exist. Specifically, the total cost, total carbon emissions, total energy consumption and total water utilized were the lowest metrics with the successful policy. However, the policy had a high service level agreement (SLA) violation rate, but it is negligible when considering the other results and real-life business adapting to the energy transition.