CAREER: Carbon Footprint Modeling and Elastic Caching for Greening Services

点击翻译按钮获取中文摘要

As Internet services scale, their growing energy and carbon footprints present conflicting challenges. On one hand, services must keep energy costs low. On the other, they should make costly investments to undo environmental harm caused by their energy footprint. Carbon offsets are transferable certificates that undo greenhouse gas emission, even when offset holders are located in carbon-heavy regions, making them attractive to large-scale, geographically distributed services. This research project lays the foundation for greening services, a new class of Internet services that buy carbon offsets for user requests routed through their servers (i.e., a service that makes other services green). People who use popular large-scale services could undo their carbon footprint by simply routing their requests through a greening service. The greening service would manage costs.The intellectual challenge for greening services is to model or confirm carbon footprints for servers outside of their control. The key insight is that emerging trends within cloud computing, e.g., energy-efficient servers, auto scaling, and open source software, provide uniformity. Dissimilarity between services, in terms of response times and energy footprints, is increasingly due to service-specific features. We use black-box machine learning approaches to infer these features.Beyond greening services, the proposed research will help system managers identify performance bugs, especially costly bugs that shift energy consumption toward datacenters with high energy costs. As part of the proposed research, the PI will conduct outreach to underserved institutions and to local, Columbus, OH, area high schools.