【作者】 熊曾刚；

【导师】 杨扬；

【作者基本信息】 北京科技大学 ， 计算机应用技术， 2009， 博士

【摘要】 网格计算和P2P计算都是作为解决大规模地理分布资源共享而新兴起来的下一代计算技术，都是目前研究的热点领域。当前大多数网格系统是中等规模的,他们一般采用集中式和分层式管理方式,虽然基础设施比较完善，但是资源管理缺乏有效地扩展；而P2P系统主要应用于互联网边缘资源的共享，如PC机的文件共享。虽然P2P系统缺乏严格的基础设施,但是P2P采用完全分布式的资源管理模式,资源可更动态地加入和退出。正因为如此,网格计算和P2P计算有越来越多的共同点，将P2P技术的一些优点应用到网格系统中,可以改进传统网格的可扩展性和动态性。然而，将P2P技术集成到的网格环境中，带来了一些新的问题和挑战，主要面临集成P2P模式的网格资源管理体系结构、资源的组织定位机制、任务调度机制以及资源管理的安全性等研究难题。针对上述问题，本文基于进程代数、Agent、Petri网等理论与技术，重点研究了集成P2P模式的网格资源发现模型、资源发现算法以及任务调度机制等问题，主要研究成果如下：1)提出了一种集成P2P模式的网格资源管理体系结构，并针对此体系结构提出了一种集成Agent技术和语义的网格资源发现模型。该模型克服了传统的集中式资源发现机制的弊端，用语义的方法进行资源匹配，能够提高资源匹配的准确性，优化网格资源发现。2)结合集成P2P模式的网格系统特征，利用Pi-演算建立起形式化模型，并借助Pi-演算工具MWB，对提出的P2PGrid服务系统实例进行行为推演及验证。结果表明，基于Pi-演算的集成P2P模式的网格系统具有活性和安全性，同时能够有效满足网格节点的服务需求。3)针对集成P2P模式的网格资源管理模型，提出了一种集成遗传和蚁群算法的混合算法来解决集成P2P模式的网格资源发现问题。混合算法利用了遗传算法和蚁群算法的优点，首先用遗传算法找到P2PGrid的资源粗集,然后再用蚁群算法求资源的精确解,并与传统的洪泛算法进行了比较，证明其性能更好。4)根据集成P2P模式的网格任务调度的特性，首先利用层次颜色Petri网来描述集成P2P模式的网格任务调度模型，并根据此模型提出了一种两阶段的网格任务调度算法。然后使用层次颜色时延Petri网来描述两阶段网格任务调度,利用Petri网的可达任务图构造算法，对Petri网的性能进行了分析。最后，通过一个简单的实例进行了分析，得到网格任务的最佳调度方案。最后在ChinaGrid的CGSP基础上构建了一个P2P模块，初步实现了一个集成P2P模式的网格环境。然后利用所设计的网格资源监控与分析系统(Grid Resrouce Vision andAnalysis System，GVAS)对本文提出的遗传蚁群混合算法进行了验证，取得了较好效果。更多还原

【Abstract】 Grid computing and P2P computing emerge as the next generation of computing technologyfor a solution of sharing the large-scale geographical distribution resource. They are both hotresearch topics in recent years. Most of grid systems are medium-scale at present, and they aregenerally applied centralized or hierarchical mode to managing resource. Although there areperfecting infrastructures, grid resources management still lacks effective scalability. On theother hand, P2P systems are mainly used in sharing the edge resources of Internet, such as file-sharing of personal computer. Although the lack of strict infrastructure, P2P systems adopt afully distributed resource management mode in which resources could be joined in and outmore dynamically. Therefore, the grid computing and P2P computing have more and morecommon ground. P2P technology, the advantage of which is applied to grid system, couldimprove the scalability and dynamicity of the traditional grid.However, P2P technology integrated into the grid environment brings about some newproblems and challenges. The main research problems we are faced with are the grid resourcemanagement architecture integrated P2P mode, the mechanism of resource organization andlocation, task scheduling mechanism, security of resource management, and so on.Aiming at the above issues, this dissertation focuses on grid resources discovery modelintegrated of P2P mode, resources discovery algorithm and task scheduling mechanism basedon the theories of Process Algebra, Agent, Petri nets.The main research contents aresummarized as follows:1) First, grid resource management architecture integrated P2P mode is put forward. Then,based on this architecture, we give an grid resource discovery model integrated agenttechnology and semantics which overcomes the shortcomings of the traditional centralizedresource discovery mechanism, using semantic approach for resources matching to enhance theaccuracy of resources matching and optimize the grid resources discovery.2) According to grid system features integrated P2P mode, a formal model is establishedusing Pi-calculus. The example of P2PGrid service system is deduced and verified with the helpof Pi-calculus tools MWB. The results show that the grid system integrated P2P mode, which is based on the Pi-calculus, has flexibility and security, and effectively meets the service demandof grid nodes.3) Based on grid resource management model integrated P2P mode, a mixed algorithmintegrated genetic algorithm and ant algorithms is proposed to solve the problems of gridresources discovery integrated P2P mode. The mixed algorithm combines the advantages ofgenetic algorithm and ant algorithms. Firstly, the genetic algorithm is used to find the resourcecoarsest of P2PGrid, then through the ant algorithm to find accurate resources. What’s more, thealgorithm is compared with the traditional flooding algorithm and the result proves that it canimprove the performance of resource discovery.4) According to features of grid task scheduling integrated P2P mode, firstly grid taskscheduling model integrated P2P mode is described by a Hierarchical Colored PetriNets(HCPN). Based on the model, we propose a two-phase grid task scheduling algorithm, andapply Colored Timed Petri Net to describing the two-phase grid task scheduling algorithm.What’s more, The performance of Petri Net is analyzed by reachable task graph constructionalgorithm of Petri Net. In the end, we analyze a simple example and give a best grid taskscheduling plan.Finally, we construct a P2P module based on CGSP of ChinaGrid and realize a gridplatform integrated P2P mode. At the same time, Grid Resource Vision and Analysis System(GVAS) is designed and accomplished under this grid environment. Based on the above, theGAA algorithm is carried out on the platform, and GVAS monitors the whole process. Theresults of analysis agree with the conclusions of the paper.更多还原