DESC: Type I: Enabling Carbon-Zero Colocation Data Centers via Agile and Coordinated Resource Management

多租户托管数据中心（也称为主机托管）是支持蓬勃发展的人工智能产业和数字经济的关键基础设施。数据中心数量和规模的快速扩张导致了巨大的能源需求。虽然托管运营商和租户都渴望实现碳可持续性并最终实现“零碳”，但将这一愿景变为现实面临着重大挑战。鉴于租户在主机托管中的孤立服务器管理，该项目旨在回答，如何重新设想综合能源需求，以利用间歇性的可再生能源供应来减少碳足迹？中心如何动态配置租户的服务器以实现运营碳效率，同时解决对服务器寿命的负面影响，以避免增加隐含碳足迹？而且，如何以灵活的方式调节租户的服务器功耗，以细粒度跟踪间歇性可再生能源？为了解决这些具有挑战性的问题，本项目建议开发（1）一种计算效率高的机制，以协调间歇性绿色能源分配给不同的租户;（2）一种基于学习的算法，以动态地提供租户的服务器，以最小化运营碳足迹，同时保持较低的体现碳足迹;以及（3）灵活的服务器功率调制框架，其将硬件功率旋钮（DVFS、睡眠状态）与软件功率旋钮（多功率二进制）相结合，以实现更快和更大的功率自适应，从而跟踪绿色能量可用性。该项目对工业界和科学界产生了广泛的影响。为了保持环境可持续和健康的数据中心行业的发展势头，该项目将优化所有者运营的数据中心的探索扩展到数据中心的一个关键和未开发的部分-多租户托管数据中心，其中租户的服务器管理是孤立的。它可以促进未来数据中心管理的转变，并最终改变人工智能行业和数字经济的发展方式。该项目还包括一个重要的教育组成部分，并提供了大量的机会，培养和吸引学生从代表性不足的群体从事计算机科学。研究成果将转化为现有课程。此外，利用PI机构中大量的少数民族学生，该项目将对少数民族学生的教育和职业产生深远的影响。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

Multi-tenant colocation data centers (a.k.a. colocation) are critical infrastructures to support the booming artificial intelligence industry and digital economy. Rapid expansion in both number and scale of data centers results in formidable energy demand. While both colocation operators and tenants are eager to achieve carbon sustainability and ultimately "carbon-zero", turning this vision into reality faces significant challenges. Given tenants' siloed server management in a colocation, this project aims to answer, how can the combined energy demands be reimagined to make use of the intermittent renewable energy supply to reduce the carbon footprint? How can centers dynamically provision tenants' servers for operational carbon efficiency while addressing the negative impact on servers' lifespan to avoid increases in embodied carbon footprint? And, how can tenants' server power consumption be modulated in an agile manner to track intermittent renewables at a fine granularity? To address these challenging questions, this project proposes to develop (1) a computationally-efficient mechanism to coordinate the allocation of intermittent green energy to different tenants; (2) a learning-based algorithm to dynamically provision tenants' servers to minimize the operational carbon footprint while keeping the embodied carbon footprint low; and (3) an agile server power modulation framework that combines hardware power knobs (DVFS, Sleep states) with software power knobs (multi-power binaries) to enable faster and greater power adaptation to track green energy availability. This project has broader impacts on the industry and the scientific community. To keep the momentum for an environmentally sustainable and healthy data center industry, this project extends the exploration of optimizing owner-operated data centers to a critical and under-explored segment of data centers --- multi-tenant colocation data centers where tenants' server management is siloed. It can catalyze a shift in the management of future data centers and ultimately transforms the way that the artificial intelligence industry and digital economy evolve. This project also includes a significant educational component and provides abundant opportunities to nurture and attract students from under-represented groups to engage in computer science. The research results will be transitioned into the existing curriculum. Further, leveraging the large population of minority students in the PI's institution, this project will bear a profound impact on the education and careers of minority students.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.