MP-DQN Based Task Scheduling for RAN QoS Fluctuation Minimizing in Public Clouds

Cloud network integration (CNI) has been a new paradigm to better support diverse vertical applications. The virtualized mobile network deployed in private and public clouds is regarded as the trend of future network evolution. However, it is challenging for radio access network (RAN) protocols to be deployed in public clouds because of the strict requirements for stable cloud resources. In a CNI environment, there coexist a large number of services (e.g. network services and cloud services) and frequent task scheduling will result in a great deal of resources fluctuation, thus degrading RAN performance. To the best of our knowledge, current researches in CNI interests ignore the high processing requirements of RAN. Therefore in this paper, we propose a multi-pass deep Q network (MP-DQN) based short term task scheduling strategy to minimize the quality of service (QoS) fluctuation of RAN deployed in public clouds. First, taking into account the differences in the relationships between resources and QoS among various services, we formulated a continuous decision problem of task scheduling. Then, We employ MP-DQN to solve the decision problem, jointly optimizing the services QoS and the task scheduling success rate. We conduct a real-world experiment to obtain the cloud RAN CPU-QoS model. The experimental results reveal that our proposed MP-DQN based task scheduling strategy performs significantly better in minimizing RAN QoS fluctuation than the conventional task scheduling strategy.