Built at numerous distributed locations in proximity to end users, edge data centers are essential ingredients of the next-generation computing paradigm to enable ultra-low latency- for emerging services such as augmented reality and assisted driving. Because of high capital expenses and/or limited physical space, however, the infrastructure capacity of edge data centers is often severely constrained and difficult to scale up for accommodating the constantly growing demand. To overcome the capacity constraint without compromising data center availability, this project exploits two important resources --- distributed energy storage devices (such as battery units) and cold air thermal energy storage --- and transforms them into server sprinting enablers for delivering the best service quality. The key contribution of this project is an intelligent and scalable framework based on state-of-the-art reinforcement learning, which holistically coordinates servers' sprinting decisions on the fly for maximizing the system-wide performance in highly dynamic and practical environments. This project considers two complementary types of edge data centers: those where the physical infrastructure and servers are managed by a single entity; and those where different tenants house their own servers and the physical infrastructure is controlled by the data center operator.
This project can push the performance beyond the currently achievable limit in edge data centers, translating into better service quality and higher user satisfaction without costly infrastructure capacity expansion. It extends the exploration of computer system optimization to a new paradigm where scalable online learning is the core, and thus can catalyze a shift in future system designs. This project also contains a significant educational component and provides broad opportunities to attract a diverse population of 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.
