An Adaptive Learning Approach for Efficient Resource Provisioning in Cloud Services

摘要

The emerging cloud computing service market aims at delivering computing resources as a utility over the Internet with a high quality. It has evolving unknown demand that is typically highly uncertain. Traditional provisioning methods either make idealized assumption of the demand distribution or rely on extensive offline statistical analysis of historical data. In this paper, we present an online adaptive learning approach to address the optimal resource provisioning problem. Based on a stochastic loss model of the cloud services, we formulate the provisioning problem from a revenue management perspective, and present a stochastic gradient-based learning algorithm that adaptively adjusts the provisioning solution as observations of the demand are continuously made. We show that our adaptive learning algorithm guarantees optimality and demonstrate through simulation that they can adapt quickly to non-stationary demand.