RINGS: Resilient Edge Networks with Data-driven Model-based Learning

点击翻译按钮获取中文摘要

Next-generation cellular wireless edge networks with embedded computational capabilities and services have become a critical infrastructure that enables secure, robust, and high-performance applications in many domains, including education, business, transportation, healthcare, and entertainment. Meanwhile, the availability, reliability, and resiliency of edge networks are constantly being challenged by both the expected resource and demand variations, e.g., diurnal application traffic patterns, user mobility, and random link/node failures, as well as the unexpected level-shifts of operating conditions, e.g., traffic flash-crowds triggered by emerging events, major infrastructure failures after coordinated malicious attacks and natural disasters. The project is developing novel hybrid learning solutions for autonomous and resilient wireless edge networks from the angle of joint provisioning, allocation, and scheduling of communication and computation resources. It includes several research thrusts: 1) To achieve high efficiency and robustness in the face of expected demand/resource variations, the project team is investigating robust communication and computation resource provisioning at long time scales. At short time scales, the team is studying adaptive routing under the Model Predictive Control framework coupled with Adaptive Dynamic Programming (ADP). The researchers are also investigating adaptive scheduling of virtual middle-boxes using a domain-knowledge enriched reinforcement learning framework. 2) To recover from major disruptions, the project team is studying how to progressively bring network services back by strategically utilizing backup and external resources. Resilient progressive recovery is achieved through joint provisioning, routing, and scheduling based on hybrid online learning. The stability of the recovery process is analyzed under the Robust ADP framework. 3) Individual research components are being evaluated using mathematical analysis, trace-driven simulations, and experiments on the CloudLab testbed. The project is generating new theories and algorithms on hybrid learning that can be used to study complex systems in other domains. Valuable research opportunities are being created for graduate and undergraduate students, especially women and 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.