Research On Dynamic Resource Scheduling For PaaS Platform

In order to satisfy the dynamic resource requirements of customers, cloud computing virtualizes physical resources in datacenter with virtualization technology, and makes resources as a pool of services. Since its appearance, cloud computing has developed rapidly, and been widely used in various industrial applications due to its low cost, high reliability, high scalability characteristics. To meet the requirements of massive resources, cloud computing providers will deploy data centers around the world. Then, a mass of physical resources gathers in each data center and different layers of cloud computing further increase the complexity. Therefore, it becomes a challenging task to design an efficient and reasonable resource scheduling mechanism for a cloud environment.Most of the existing studies of cloud computing resource scheduling method focus on energy consumption optimization of IaaS (Infrastructure as a Service) layer. This thesis is committed to solve the resource scheduling, load balancing, load prediction, and instance auto-scaling problems, based on the PaaS layer, especially the private PaaS platform.Firstly, this paper investigates existing resource management strategies of different PaaS platforms, especially Cloud Foundry (CF), which is a typical open source PaaS platform. This paper also analyses the architecture and design of CF in detail.Secondly, this paper studies several models which can be used to predict the cloud application resource usage, including ARIMA (Autoregressive Integrated Moving Average) models, ANNs (Artificial Neural Networks), and a hybrid model that combines both ARIMA and ANN models.Then, based on the prediction results, a dynamic resource scheduling algorithm is proposed, and experimental evaluations prove that the proposed algorithm has high real-time performance and can indeed improve the resource utilization of PaaS layer as well as the QoS (Quality of Service) of cloud applications.Next, this paper implements a new DEA (Droplet Execution Agent) component that enriches CF with many new features, such as dynamic resource scheduling, instance auto-scaling and load balancing of DEAs.Finally, several experiments are designed to verify the effect of the improved CF platform. These experiments show that the improved platform can effectively improve the resource utilization and QoS of cloud applications.