CAREER: Data-Driven Hardware and Software Techniques to Enable Sustainable Data Center Services

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Enabling environmentally sustainable data center systems while supporting escalating computing needs has emerged as a “grand challenge” of our times. Unfortunately, reducing emissions is especially challenging for modern data center applications, since their design often prioritizes performance and scalability, causing increased emissions from (1) performing extraneous work and (2) requiring more computing hardware. To reduce emissions while achieving performance and scalability goals, this project will leverage data-driven techniques to radically rethink the cross-layer hardware-software data center computing stack from the ground-up, with carbon efficiency at the forefront. This project also includes a comprehensive education and outreach plan that is seamlessly integrated with the research. The highlights include: (a) adding new sustainability modules in high school, undergraduate, and graduate curricula; (b) conducting sustainability-focused workshops and tutorials at academic conferences; and (c) engaging students groups in sustainability research. Thus, this project’s direct impact is redefining how future data center systems are designed, managed, and taught in a sustainability-first manner. Aiming at reducing emissions caused by microservices, this project systematically studies microservices through a sustainability lens. The project’s key innovation is a radically redesigned hardware-software systems stack with carbon-aware hardware and novel software systems that leverage such hardware while achieving microservices' scalability goals. On the hardware front, this work will take a new approach using the sustainability principles of reducing and reusing to systematically design carbon-aware hardware at the server, device component, and microarchitecture levels. First, this work will identify when to reuse older servers to run latency-tolerant microservices. Second, to reduce devices, this work will develop a tool to only accelerate key common primitives and analyze resource wastage to design reduced components. Third, this work will design leaner microarchitectures. On the software front, this work will develop a novel carbon-aware data-driven resource management framework to enable the carbon-efficient designs at each hardware level. By pursuing data-driven techniques, this work will highlight the value of using data centers' always-on telemetry to achieve its designs. This cross-disciplinary research cuts across three areas: computer architecture, software systems, and new application paradigms, opening new opportunities for interdisciplinary work on sustainable computing. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.