The Key Technology Research And Implementation Of Numerical Weather Prediction In Cloud Computing Environment

With the continuous development of numerical weather prediction technology, numerical weather prediction requires a lot more high-performance computing resources.Compared with traditional numerical weather prediction,ensemble prediction adds the method of generation of the initial perturbations to improve prediction effect.Because each model prediction of initial samples is independent, it is fit to accomplish under web environment in distributed way . So,under distributed environment using resource aggregation can improve timeliness of ensemble prediction.Cloud computing based on distributed computing, resource aggregation and virtual computing can solve problem ensemble prediction faced. Service-oriented computing resources is very important manifestation form of cloud computing.Cloud computing can provide resource including application programs and infrastructures by web service,and share aggregated resource to every user using virtualization.The user can use various terminal equipment to access cloud computing service through Internet.Ensemble prediction in Cloud Computing Environment faced two problems.One of them is the schedule problem how to distribute the initial samples which ensemble prediction produces to each cloud node to reach purpose of time and cost balance optimization.Traditional ensemble prediction systems use the form of assigning node manually and staticall or sequential execution in single node.This method decreases the timeliness of ensemble prediction and can not adapt the dynamic change of node resource.Another is the access interface problem how to provide ensemble prediction service by visual and friendly form to promote ensemble prediction development and enlarge application range. Traditional ensemble prediction systems run application program by command line,and involve a series of complex process. It has high threshold, highly professional and poor interaction. So it is difficult for normal user to access service by commad line,and restrict ensemble prediction development.Considering the problems which ensemble prediction in cloud computing environment faced and combining with cloud computing technology, in this paper we design and implement ensemble prediction in cloud computing environment.First,desig- n and analyse the requirement,function,structure,workflow and design pattern of the system to introduce whole frame of ensemble prediction system. From the standpoint of requirement and function, analyse applicability and advantage of ensemble prediction system in cloud computing environment.In terms of structure,component,workflow, analyse function feature and relationship of each module,and introduce system design pattern through MVC design pattern.Around ensemble prediction scheduling problem, a WSRF-based schedule system is implemented with the computing features of model prediction in ensemble prediction,which can be dynamically adaptive to cloud environment.One kind of scheduling algorithm based on time and cost of ensemble prediction application balance optimization is proposed,which can be adaptive to the dynamic changes of cloud nodes and assign the optimal number of CPU for model prediction task dealing with initial samples. The schedule system can save computational costs as much as possible,while shorten the time of prediction.It can improve the time effectiveness of the ensemble prediction which has the characteristic of large-scale computing,and effectively save the computing cost of ensemble prediction.To solve access interface problem,we design and implement ensemble prediction cloud portal(NPCPortal) including the modules of ensemble prediction and cloud computing.NPCPortal can provide the web interface for users to access cloud resource and ensemble prediction service to solve the problem that poor interaction of the traditional ensemble prediction system. Implement user rights management for NPCPortal by using a double-layer access control technology. In portal presentation layer,design a plug-in called domainselect to provide visual setting of forecast area parameters.In terns of implementation of ensemble prediction system,we deploy cloud environment based on Nimbus.Nimbus supports WSRF and EC2 protocols,and WSRF provides development mechanism of the stateful web services.The System can define and manage the status of initial sample data by resource class.Implement NPCPortal based on Gridsphere and each components of NPCPortal by portlet.