Research On Hybrid Cloud Scheduling Models And Algorithms For Optimizing Cost Of Application Service Provider

Cloud computing, with its infinite resource and pay-per-use cost model, has become the hot point both academia and industry with wide attention and discussion. Meanwhile, with the development of public cloud service provider, more and more application service provider are considering abandoning the thought of building their own data centers and booking public cloud resource to service. Currently, there is a trend to take advantage of the virtualization technology and cloud computing and expand the service into largescale and wide-range situation under higher service quality and lower cost.According to the deployment mode, cloud computing system can be divided into four kinds, public cloud, private cloud, community cloud and hybrid cloud. Private cloud has the advantage of good data security and good manageability, buts poor in cost. Public cloud, on the other hand, has low booking cost and also low security. Hybrid cloud system can combine the advantage of both and further improve the quality of serviceand reduce the costThis thesis standing at the role of application service provider using the hybrid cloud model to provide the service, focuses on the scheduling model and related algorithm to achieve the minimum cost under a certain kind of quality of service. Firstly, a booking cost scheduling model is proposed according to the public cloud billing model and character of user service requests. Secondly, with the model, an algorithm framework is proposed. In first step, the future period of user request is forecasted according to the statistic data. Then, using longterm scheduling algorithm to achieve the optimal solution using the forecasted daa. Short-term scheduling algorithm is also used to do autoscaling using monitoring data. Thirdly, a hybrid cloud system based on OpenStack and AWS is designed. After that, more detailed forecast algorithm, the long-term and short-term scheduling algorithms are proposed based on the content delivering application. Finally,the validation of the algorithm is verifiedby combining both real world data and simulation.