【作者】 郭东；

【导师】 殷保群；

【作者基本信息】 中国科学技术大学 ， 控制理论与控制工程， 2011， 博士

【摘要】 随着信息技术的飞速发展,各种新颖的基于计算机网络的应用层出不穷,P2P网络就是其中典型的代表。虽然以往的研究也提出了许多P2P网络模型和关键技术,但对P2P网络动力学建模方面的研究却为数不多。同时,在P2P网络的应用领域目前存在着大量的问题,而其中很多问题可以在现有的应用支撑环境下采用现代控制理论和系统分析的方法获得解决。本论文旨在面向P2P网络这样一个应用广泛的对象,研究其动力学建模和一些关键算法,以便为实际网络系统的应用提供有效的解决方案和指导性建议。本文的主要工作和创新性表现在以下几个方面：1.研究了P2P文件共享系统的动力学建模和一些关键算法。我们以节点之间待传输的数据量作为系统的状态,提出了一种用确定性微分方程组来描述系统状态变化的动力学模型。该模型不仅考虑了影响节点数据传输速率的网络带宽,拷贝份数,存储空间,还考虑了P2P网络中节点选择算法,带宽分配算法,激励算法等因素。通过分析P2P文件共享系统中的算法对节点状态的影响,我们导出了方程的具体形式。实验结果表明该模型能够反映P2P文件共享系统的本质特征,从而为设计高性能P2P网络,改进现有算法,以及研究P2P网络的稳定性提供了理论基础。2.运用经济学中的市场机制原理,为P2P网络建立了一种基于动态价格的激励模型(Dynamic price-based incentive model, DPIM),有效地提高了P2P网络的可用性。在此模型基础之上,我们主要研究了其中的激励算法。仿真和分析结果表明,此种供求关系决定带宽价格的激励模型可以充分利用市场自主调节的功能,促使节点增加上传带宽和共享资源,避免了搭便车行为,提高了系统效率,打击了作弊行为。3.研究了媒体分发网络MDN(Media Delivery Network)系统中对等节点PN(Peer Node)的关键算法。在P2P文件共享系统的动力学模型基础上,定义了MDN的系统状态,并为MDN系统建立了动力学模型。该模型考虑了PN上传速率,PN存储空间大小,PN节点的部署算法,节点选择算法,带宽分配算法等因素对状态的影响。为了提高MDN系统的服务性能,我们还提出了基于状态的改进算法。仿真结果表明,所建立的模型能够准确反映MDN系统的本质特征,为系统的性能优化提供了一种方法。更多还原

【Abstract】 With the development of information technology, many novel applications of computer networks have appeared. P2P is one of the most important applications. Although there were many models for P2P networks, few have addressed the model of dynamic behaviors. Meanwhile, P2P networks have a lot of problems, but modern control theory and system analysis methods provide a new perspective in solving these problems under existed application support environment. To this end, we study the dynamical modeling for P2P networks and some key algorithms. Then, we provide effective solutions and some advances for actual P2P networks.The main works and innovation are as follows:1. We propose a dynamical model for P2P file sharing systems and study some key algorithms. States of peers are defined as the amount of data which will be transmitted. We formulate a set of deterministic differential equations to describe the evolution of systems states by taking network physical characteristics (the physical upload, download bandwidth, storage space, etc) and software protocols (bandwidth allocation algorithm, peer selection algorithm, peer incentive mechanism, etc) into consideration. Then according to particular algorithms, we give the corresponding concrete models. Experiments show that the proposed model can adapt to different P2P file sharing systems. Therefore it provides a theoretical basis for designing high-performance P2P networks, improving existing P2P algorithms and researching stability of the P2P systems.2. We study the market mechanisms of economic principles and utilize it to model P2P file sharing systems. We name it as dynamic price-based incentive model (DPIM). Then, we study the incentive algorithm of DPIM. The results of simulation and analysis indicate the model can reduce free riding in P2P networks, punish cheating behaviors, and enhance file sharing efficiency and service quality.3. We study some key algorithms in the P2P-based Media Delivery Network (MDN). Based on the dynamical model of P2P, we define the states for the MDN and provide a novel dynamic model to character the behaviors of MDN systems. PN upload bandwidth, storage space, storage allocation algorithm, bandwidth allocation algorithm and PN selection algorithm are considered in the model. In order to improve the performance, some state-based algorithms are proposed. The results of simulation show that the model captures the characteristics of MDN in a more accurate way and provides a method for dynamic optimization of the system.更多还原