云服务中间件的研究及其在台风预报系统中的应用

发布时间：2018-04-14 16:25

本文选题：云服务 + 中间件　；参考：《福州大学》2014年硕士论文

【摘要】：在Web应用日益普及的今天,Web服务器承担的任务越来越繁重,特别是进行大数据量的计算时,服务器的硬件资源成了计算效率的瓶颈。而单纯地提高服务器本身的性能效果并不明显,而且存在资源浪费。如果能将计算任务分解成若干个子任务,分配给闲置的计算服务器,那么由于并行计算的优势,效率将大大提高。云计算技术可以解决大任务计算的效率问题。云计算是一种商业式的计算模型,它可以将计算任务分解成小任务,分布到大量由计算机构成的资源池上,使得各种应用能够根据需要获取计算能力、存储空间和信息服务。由于云计算任务的计算过程是在各台计算机的本地进行的,在网络上传输的仅有任务描述和计算结果,所以网络传输的代价小,计算效率取决于计算节点的平均性能和负载均衡算法。本文提出了一种基于多Agent的云服务中间件的框架体系,分为三个层次,用户接口层、SOA层和资源管理层。用户接口层负责任务接口的描述,Web请求端通过接口调用云服务模式下的任务计算。SOA层是面向服务的,它管理了云服务中间件向用户接口提供的服务,维护服务列表,可以根据用户的请求查找相应的服务。资源管理层是整个云服务中间件的核心部分,它包含了一个负载均衡器和监控器,提供容错服务和日志服务,实现了任务分解、分发和结果汇总的逻辑,维护计算节点列表。一个完整的Agent平台由多个Agent容器组成,每个容器下有若干个Agent和一个消息调度模块,Agent之间通过发送和接收消息进行通信。在台风预报云服务中间件CSMTF (Cloud Service Middleware Of Typhoon Forecast)中,有一个负责任务分解、分发和结果汇总的Agent,称为主控节点。主控节点接收从请求端发来的计算任务,内置的负载均衡器实时获取各台闲置计算机的性能信息,将任务分配给轻载节点。主控节点内置的日志监控器负责各Agent日志的收集和记录,以便管理人员查看了解Agent的工作情况。同时,主控节点还内置了容错监控器,检测Agent运行状况,恢复出错的Agent。云服务中间件中有若干个负责子任务计算的Agent,称为计算节点。计算节点实时发送本机的负载信息给主控节点,接收来自主控节点分发的子任务,在本地计算,并向主控节点返回计算的结果。计算节点也带有一个日志监控器,用于记录本地Agent的信息,内置的容错监控器用于本地恢复。台风预报系统是一个典型的Web应用,它是一个会商式的群决策支持系统(GDSS)。其中的专家预报路径就涉及了大数据量计算,Web服务器需要对预测的路径与数据库中的所有历史台风路径进行相似度计算,并且要求实时显示最相似路径。而数据库中的路径点数有成千上万,在Web服务器资源有限的情况下,很难进行有效的计算。本文将云服务中间件的技术应用到台风预报系统上,以提高相似路径计算的效率。最后,本文搭建实验环境对CSMTF系统做了性能测试,与Web服务器单机计算相比,计算速度有显著的提高,大大缩短了服务器的响应时间。
[Abstract]:In the Web application is becoming more and more popular, Web server takes more and more arduous task, especially the calculation of large amount of data, the server hardware resources has become a bottleneck of computational efficiency. And simply improve the performance of the server itself is not obvious, and there is a waste of resources. If it can be decomposed into several computational tasks sub tasks, assigned to the computing server idle, then because of the advantage of parallel computing efficiency will be greatly improved. Cloud computing technology can solve the problem of efficiency calculation tasks. Cloud computing is a business model type, it can compute tasks into small tasks, distributed to a large number of resources by computer then, according to the needs of various applications that access computing power, storage space and information services. The calculation process of cloud computing task is performed in the local computer, The only task in the transmission network description and calculation results, so the network transmission cost is small, the computational efficiency depends on the average performance of computing nodes and load balancing algorithm. This paper proposes a framework of cloud service middleware based on multi Agent, divided into three levels, the user interface layer, SOA layer and resource management the user interface layer. The interface layer is responsible for the task description, the Web request by calling the interface end cloud service model under the task of calculating the.SOA layer is service oriented, it manages the cloud service middleware provides to user interface services, maintenance services list, according to the user's request to find the corresponding service resource management layer is the core part. The cloud service middleware, which includes a load balancer and monitor, provide fault tolerance service and log service, realize the task decomposition, distribution and results summary logic, maintenance plan A list of nodes. A complete Agent platform is composed of a plurality of Agent container, and a plurality of Agent and a message scheduling module of each container, Agent communicate by sending and receiving messages. In the forecast of typhoon cloud service middleware CSMTF (Cloud Service Middleware Of Typhoon Forecast), a responsible for task decomposition a summary of the results of the distribution, and Agent, known as the master node. The master node receives a request sent from the terminal computing tasks, load equalizer performance information real-time access to each idle computer built-in, assign tasks to light load nodes. The master node built-in monitor log log Agent is responsible for the collection and recording. In order to understand the management view Agent work. At the same time, the master node has built-in fault-tolerant monitor, testing the running condition of Agent, middle Agent. cloud service part error recovery in if A responsible for sub task computation Agent, called compute node. Calculation of real-time transmitting node of the machine load information to the master node, the master node receives from the distribution of sub tasks in the local computation, and return to the main control node calculation results. The compute nodes with a log is used to record the monitor. The local Agent information monitor built for local fault tolerant recovery. Typhoon forecast system is a typical application of Web, it is a type of consultation group decision support system (GDSS). The expert forecast path involves the calculation of a large amount of data, the Web server needs all the historical path and database on the prediction of typhoon in the similarity calculation, and real-time display the most similar path. The path points in the database there are tens of thousands of, in the Web server under the condition of limited resources, it is difficult to effectively calculate the. The cloud service middleware technology is applied to the typhoon forecasting system, to improve the efficiency of similar path calculation. Finally, this paper set up the experimental environment of the CSMTF system performance test, compared with the Web single server computing, the computing speed is significantly improved, greatly shorten the response time of the server.

【学位授予单位】：福州大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【相似文献】