【作者】 文元桥；

【导师】 余胜生；

【作者基本信息】 华中科技大学 ， 计算机系统结构， 2006， 博士

【摘要】 随着人类对自然规律认识的日益深入和人类科研领域的不断扩展,现代科学研究呈现出大规模、多学科交叉的特点,这也是现代科学计算的一个发展方向。和传统的单一学科的大规模科学计算相比,大规模多学科交叉的科学计算更强调多个学科的协同与合作。这种协同合作不仅仅有不同科学计算程序之间的协作,也包括人和人之间的协同合作。在这种情况下,传统的面向某一特定科学领域的科学问题求解环境已经不能很好的满足现代科学计算的需求。因此,需要设计一种更强调协同和合作的科学问题求解环境——协同科学计算环境(CSCE: Collaborative Scientific Computing Environment)。协同科学计算环境旨在为多学科的科学计算用户提供一个协同工作的虚拟空间,它向用户提供方便、智能的应用界面和丰富的协同功能,使得不同学科的用户能够透明的享受高性能计算,实现对各自领域科学问题的计算机求解,也可实现对复杂问题求解过程中的跨学科的协同合作。在协同科学计算环境中,不同学科的用户不仅仅可以便捷、高效的获取包括科学计算程序、应用程序、科学数据、文档、高性能计算机等计算资源,而且可以利用系统提供的科学计算试验管理工具帮助他们实现对复杂科学问题的定义、管理、求解、分析,从而使得他们能有更多的时间和精力来关注自己的研究领域,而不是关注如何计算本身。以建立区域气候模拟系统为研究原型,探讨利用Peer-to-Peer技术和移动Agent技术构建协同地球科学计算环境(GeoCoEn: Geophysical Collaborative Computing Environment)的一些关键技术。其目的在于为开展区域气候模拟研究的不同学科的研究人员提供一个易用、透明、开放的高性能协同科学计算环境,满足分布在不同地方的用户开展协同模拟、共享资源、共享计算的需求,探索构建以网络为基础的协同科学计算环境的理论和技术要求。根据GeoCoEn体系结构的需求分析,吸取近年来问题求解环境(PSE: Problem Solving Environment)和协同问题求解环境(CPSE: Collaborative Problem Solving Environment)体系结构和应用框架方面的优秀思想,GeoCoEn基础体系结构采用了基于P2P和多Agent集成的多层体系结构,满足了GeoCoEn对开放性、灵活性、可扩展性、可移植性以及透明性等方面的需求。P2P和多Agent的集成既能够充分发挥P2P在对等计算、资源共享、资源发现以及协同工作方面的优势,也能发挥移动Agent既能够保证系统良好的伸缩性、健壮性,也能克服系统的异构性,还能够实现高效的软件重用以及提供灵活多样的协同合作方式等优点。在GeoCoEn中,每个用户就是一个对等体——GeoPeer,而GeoPeer之间通过采用基于社区的混合P2P结构实现协同合作。这种组织结构既能准确的反应出每个Peer的个体特征,也能体现其团体特征,同时也能够保证Peer和Peer之间能够准确、迅速的相互感知和定位,提高了系统在资源定位、文件搜索等方面的效率。GeoPeer提供的包括Peer管理服务、数据管理服务和通信服务等在内的多种服务为实现分布式环境下的协同合作及大规模、广泛的数据与资源共享提供了良好的支持。在分析和讨论多模式耦合技术和科学需求的基础上,根据GeoCoEn基础体系结构的基本思想,利用多Agent技术实现了分布式环境下的多模式耦合,构建了一个具有开放架构的分布式多模式耦合模拟系统(DMCSS: Distributed Multi-model Coupling Simulation System)。根据GeoCoEn科学活动低重复性和高动态性的特点,建立了一个树状结构的动态科学工作流模型,并利用有向图理论对模型进行了描述。基于案例推理的重用为解决科学工作流低重复性问题,实现科学工作流从单个计算步骤到整个流程定义的多层次重用提供了有效的解决手段。更多还原

【Abstract】 With the increasingly widening of mankind knows and the ever-extending of human research domains, the modern science researches present large-scale and inter-disciplinary characteristics, which is the characteristics of modern scientific computing too. Compared with the traditional large-scale single-disciplinary scientific computing, the modern large-scale inter-disciplinary scientific computing emphasizes inter-disciplinary cooperation and collaboration more, which include not only the cooperation and collaboration among different scientific applications, but also the cooperation and collaboration among individuals. Therefore, it is necessary to develop a collaborative scientific computing environment (CSCE), which emphasized cooperation and collaboration more, to take place the traditional scientific problem solving environment (PSE), which are designed for the specific domains and can not satisfy the support requirements of multi-disciplinary scientific computing.The CSCE, which has convenient, intelligent interface and multiple collaboration functions, is designed to provide a virtual collaborative place for multi-disciplinary scientists to share the high performance computing transparently, resolve the problems of their own disciplines by using the computing facility and realize the cross-disciplinary collaboration for the complex cross-disciplinary problems. Under the sharing place provided by CSCE, not only numerical models, application, scientific data, documents and high performance computing facilities can be shared conveniently and efficiently by scientists of different domains, but also management tools for scientific experiments are provided to define, manage, resolve and analyze the complex large-scale scientific problems. Thus, the scientists can focuses on the scientific problems of their own domains rather than the computing itself.By taking regional climate system modeling as the research prototype, the dissertation investigates the key problems of building GeoCoEn (Geophysical Collaborative Computing Environment) based on distributed web environment by using Peer-to-Peer and mobile agent technology. The purpose of the dissertation is to develop a convenient, transparent and open high performance CSCE for the scientists focusing on regional climate researches, and satisfy their requirements of collaborative computing, data sharing and computing sharing. Moreover, it explores the necessary theory and technology of building web-based scientific computing environments. The main contributions of this dissertation include:The architecture design focuses on how to build flexible CSCE by using on-shelf applications, system software and foundation facilities. A multi-layer architecture, including application layer, management layer, execution layer and supporting layer, is designed to support openness, flexibility, extensibility, portability, transparence and other abilities required by CSCE by integrating P2P and Multi-Agent System (MAS).Under GeoCoEn, the basic unit is a peer, which is named GeoPeer. The cooperation and collaboration of GeoPeers are realized by taking community based P2P organization structure. The services including peer management service, data management service and communication service provided by GeoPeer are the foundation for large-scale, wide data and resources sharing under distributed computing environments.DMCSS (Distributed Multi-model Coupling Simulation System) of GeoCoEn is a numerical model coupling system based on distributed computing environment by using mobile agents. The multi-layered system provides a flexible sharing workspace and an open framework for the cooperation of multidisciplinary researchers in geophysical modeling and makes it is possible for the researchers to implement the numerical model coupling by plugging in or pluggling out the models.A dynamic workflow model based on tree structure is designed considering the lower repeatability and higer dynamic property of sicientific workflow. It describes the detail of the dynamic workflow model by using directed graph theory. The case-based reasoning reuse provides an effective method to reuse the scientific workflow definition from separate steps to a full-process definition considering the lower repeatability of scientific workflow.更多还原