DESC:Type I: Sustainable Serverless Computing

Cloud computing has transformed our everyday lives by enabling on-demand access to powerful computer system resources from Internet-connected platforms including smartphones, digital health devices, connected vehicles, and smart home systems. Serverless computing is an emerging paradigm that enables fine granularity function-level cloud microservices to be accessible by these smart platforms.

The paradigm allows cloud computing service providers to more efficiently provision their compute resources, which translates into cost savings for both developers and service providers. This in turn is expected to make digital intelligence more accessible, affordable, and pervasive in our everyday lives. However, one of the biggest outstanding challenges with serverless computing is to support function-level performance guarantees while minimizing the environmental impact of cloud datacenters that host serverless computing.

Cloud datacenters are already a major contributor to global carbon emissions, wastewater generation, and electricity use, and serverless computing will exacerbate these pressures on the global environment. This project will involve transformative research to realize sustainable serverless computing with three major thrusts that will be addressed in an integrated manner: 1) Performance and sustainability modeling for serverless computing will be conducted to capture the performance of serverless workflows in datacenters while also characterizing carbon footprint and water use for supporting serverless computing for the first time, including the overheads from manufacturing, operation, transportation, water-use, and end-of-life decommissioning; 2) Encapsulation layer enhancements will be developed to address some of the biggest performance bottlenecks with serverless computing, such as high startup latency, low performance storage, and fault tolerance; while simultaneously minimizing operational and embodied carbon footprint associated with serverless computing; and 3) Orchestration layer enhancements will be devised based on hybrid evolutionary-learning and multi-agent reinforcement learning to minimize the environmental impact of serverless computing while meeting performance goals across geographically-distributed datacenter platforms.

This project is aligned with the goals of the National Discovery Cloud for Climate (NDC-C) program as it involves cloud-based data modeling, analysis, and optimization for a variety of emerging serverless computing applications, including weather predictions from multi-modal data and climate modeling, that promise to advance climate-related research. The project also aims to realize fundamentally more sustainable cloud computing infrastructures that can support large-scale climate science and engineering workloads in the scope of the NDC-C program while reducing the damaging environmental impacts of cloud computing datacenters that execute them.

The emphasis on characterizing and co-optimizing the performance and environmental impacts of serverless computing will improve the proliferation of low-cost cloud computing, making it more cost-effective and seamless to integrate into computing-driven services that can enrich our everyday lives. Workforce development is another important broader impact of this project, with high school students, undergraduate and graduate students being trained in the multi-disciplinary domains of high-performance computing, optimization theory, and environmental sustainability.

Close collaboration with industrial partners will also ensure timely dissemination and integration of research outcomes into real-world sustainable climate-friendly computing initiatives.

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.