Sustainable Datacenter Hardware Management

I’m currently leading a research project that is researching ways to reduce the environmental impact of datacenters, particularly centering around reducing the embodied emissions of hardware used in datacenters/hyperscale systems as well as reducing the waste generated from this hardware.

While sustainable management has focused on ways to reduce, reuse, and recycle common resources such as oil and plastics, similar strategies are understudied for managing data center server hardware, where current practice is to dispose of server hardware every 3-5 years. Minimizing hardware waste is crucial; previous work has shown that up to 50% of data center system emissions are “embodied” emissions, which result from the manufacturing and transport of server hardware. Thus, I propose that we view data center hardware resources just like other resources (e.g., oil) that we must carefully and creatively manage in an effort to build long-term, sustainable systems.

The goal of my research is to provide an innovative cross-stack solution to expose operational and embodied carbon emissions as optimization metrics in data centers and a hardware resource management strategy to minimize them. Specifically, I will research ways to practically reduce, reuse, and recycle, hardware components within data centers to amortize the carbon footprint of manufacturing server hardware and to extend their usable lifetime, leading to a carbon footprint reduction. Although I will focus on making hardware more sustainable in data centers, the overall solution I will develop will be inherently cross-stack. Specifically, reusing hardware components will lead to different generations of hardware being present in a data center. As a result, software must be aware of this heterogeneity to schedule tasks in a way that prevents performance degradation from running applications on older, and thus potentially less performance-efficient, hardware.