Prediction-Driven System Management for Robust and Green Cloud Computing Infrastructures

Dr. Xiaohui (Helen) Gu
North Carolina State University


Abstract

Cloud computing provides a new cost-effective computing paradigm by obviating the need for users to maintain complex computing infrastructures themselves. Users can lease resources from the cloud system to perform various computing tasks (e.g., big data analysis) in a pay-as-you-go fashion. However, due to its inherent complexity and sharing nature, cloud computing faces many challenges including system reliability and high energy cost. In this talk, I will present a set of prediction-driven system management techniques (e.g., online system anomaly prediction and prevention, elastic resource scaling) developed in my group to achieve robust and energy efficient cloud computing.