Carbon Accounting for Sustainable Computing in Cloud Provisioned Data Centers

Abstract

Enterprise digital operations require data-driven and scaleable solutions to monitor and manage the increasing environmental impact caused by manufacturing hardware and operating large data center warehouses. Data centers are complex systems comprising of heating, ventilation, cooling, power distribution, and workspaces to support employees managing the facility's operation. These workloads drive the global financial sectors, critical supply chains, big data analytics supporting consumer buying habits, and managing digital records for healthcare and payrolls that are critical to the function of modern society.

Although these digital operations are hidden within the confines of these large data centers, out of sight and out of mind, their impact on the environment is not negligible. The world's IP traffic is expected to exceed 2.3 zettabytes. Data centers consume significantly more energy per floor space of typical commercial buildings, 1.8% of the total energy used in the United States, and 1% of energy worldwide. This drives the need to measure and understand the impact of these workloads on the environment so that innovation and optimization can be leveraged to allow us to grow these technologies and digital opportunities sustainably.

To sustainably grow our digital presence, we need a method for tracking the environmental impact that the industry can adopt at scale while allowing for continued growth and development of digital technologies in parallel. There is interest from several government organizations to standardize a method to achieve this. This thesis attempts to illustrate a standardized and flexible way to measure the environmental impacts of digital solutions that combine the embodied emissions caused by manufacturing hardware and operational emissions from both the ICT and non-ICT systems in the data center. This solution would provide a means to track the actual ecological influence of digital operations that can be applied to information technology systems at any scale ranging from small technology systems to large cloud-provisioned data center operations. This work will focus on embodied emissions combined with operational emissions, reporting methodologies, and how that information can be used to distribute workloads using a multi-armed bandit algorithm using Thompson Sampling.

This work illustrates that embodied emissions can constitute anywhere from five to thirty percent of a server's total environmental impact. Workloads can be more than 11 times higher between workloads given the same set of hardware in a data center. Choosing the right configuration can reduce emissions by 4 times. Data center location significantly affects operational emissions, simply shifting the region where a workload is executed can reduce emissions by 24% or more. Using a data-driven approach, the spatial distribution of workloads can be implemented to optimize digital operations based on environmental objectives with as low as 30 iterations. The risk from regulation can be reduced, and a competitive advantage can be gained from implementing a sustainability-focused digital architecture.