【作者】 路卫娜；

【导师】 杨寿保；

【作者基本信息】 中国科学技术大学 ， 计算机系统结构， 2009， 博士

【摘要】 随着计算机技术和通信技术的不断发展,网络环境已经从早期相对静态的、面向特定组织和用户群体的封闭网络,转变为可公共访问的、面向大量动态用户的开放网络。在国内外,作为开放网络的典型代表,网格和对等网络成为分布式系统方向最活跃的研究领域之一。开放网络具有开放、动态、异质、实体对等自治、资源共享自愿等特性,网络的理性用户更多地表现出自兴趣和自主性,其根本目的就是最大化自身利益,而并不考虑网络的整体效用。每个节点都希望尽可能多地使用其他节点的资源,而尽可能少地贡献自己的资源,或者即使贡献自己的资源,也不保证资源的可靠性,网络中存在大量的不合作现象和欺诈行为。随着网络系统规模越来越大,资源的公平共享问题正在变得棘手,开放网络中的激励机制已经成为分布式系统研究领域的一个重要课题。本论文围绕开放网络中节点自主行为造成的不可靠的服务可用性和搭便车问题进行了研究,以期有效合理的激励理性自私的节点诚实共享自有资源,有效利用好网络已拥有的资源,提高网络综合能力。本文首先分析了开放网络的特点,对开放网络的典型代表Grid和P2P进行了概述和比较,并深入讨论了其中存在的问题,引出本文的研究内容。随后在第二章,介绍了开放网络中存在的公平共享问题,由公平性维护扩展到信誉度维护,并总结了目前存在的信誉度评价方法及信任模型;本章还分析了把市场机制应用到开放网络资源管理中的优势以及市场激励中需要考虑的关键问题。其次,针对信息不对称问题,将信誉机制引入网格市场的资源交易中,提出了基于信誉感知的资源交易机制。给资源提供者赋予一定的信誉值,以反映资源的可靠性,资源使用者可以根据信誉值判断是否交易。基于信誉感知的集合竞价机制和基于信誉感知的连续双向拍卖机制均能使资源使用者由于不可靠的服务可用性造成的损失明显减少,有效解决了信息不对称问题,两种机制各有优势。然后,为了调整自私节点的行为,提出了基于信誉的市场激励模型。假设不同的服务水平可以兑换成不同的价格,服务使用者可以根据信誉系统的推荐购买服务,最小化自己的购买风险;服务提供者模型化为具有学习能力的Agent,通过学习逐渐适应本地市场,获得优化的服务质量调整决策,最终系统将达到一个平衡。该模型能够同时保证服务双方的利益,因而能够激励节点积极地贡献和使用资源。再次,提出了P2P文件共享系统中基于惩罚的信誉模型。针对现有信誉模型存在的不足,从经济理论受到启发,将惩罚的概念引入到信誉机制的研究中,对节点的不协作行为进行惩罚,建立一种基于惩罚的信誉模型。该模型能够有效抑制节点不协作行为的发生,达到激励节点协作的目的。最后,论文第六章提出了P2P流媒体系统中积分检测相结合的激励机制。节点通过自身或子节点共享数据流获得贡献积分,贡献积分越高,节点获得数据流的优先权就越高;为了贡献积分最大化,节点有选择地对它的子节点进行检测来促进子节点数据的共享。使用该机制能解决长期历史积分问题,提高合作者比例,并维护了较高的系统性能。本文从网络动态性、节点异构性和资源的不可信性等开放网络应用呈现出的问题出发,紧紧围绕节点不可靠的服务可用性和Free-riding搭便车问题进行了深入研究,提出了网格环境中基于信誉的资源交易机制和市场激励模型,并提出了P2P文件共享系统中基于惩罚的信誉模型和P2P流媒体系统中积分检测相结合的激励机制,为解决开放网络资源共享难题展示了全新的视角和美好的前景。更多还原

【Abstract】 With the development of computer and communication technologies,network environment has changed from a close network,which is comparatively static and faced with special organizations or user groups,to an open one that can be publicly accessed and faced with many dynamic users.As the representatives of open networks, Grid and P2P have become one of the most popular research fields all over the world.Open network is open,dynamic,heterogeneous,autonomic,equal and voluntary for resource sharing.Therefore,the rational peers in the network are self-interest and try to maximize their own benefit without considering the whole utility of network. Every peer hopes to use others’ resources as many as possible,but contribute their own resources as few as possible,which leads to non-collaborative and dishonest behaviors in the network.As the network enlarges,fair sharing of the resources turns to be more intractable.Thus,incentive mechanism becomes an important issue of distributed system research area in such an open network.This dissertation discusses unreliable service usability and Free-riding problems in open networks resulting from peers’ selfishness.It focuses on the researches that can stimulate rational and selfish nodes to share resources honestly,so that the network’s integrated ability can be improved.Firstly,the dissertation summarizes the characters of open networks,describes and compares Grid and P2P which are the representatives of open networks,and presents the existing problems to elicit the research content of the dissertation. Subsequently,Chapter 2 introduces the fair sharing problem in open networks and summarizes reputation evaluation methods and trust models.This chapter also analyzes the advantages of using market mechanisms to manage open network resources,and bring forward the key of market incentive.Secondly,aiming at the severe information asymmetry problem,the dissertation introduces reputation to grid-based market resource trades and proposes reputation-aware transaction mechanisms.A reputation value is used to reflect the reliability of a resource;the users decide whether to use the resource in terms of the value.The reputation-aware aggregate market mechanism and double auction mechanism can both reduce the loss of the users resulting from unreliable service usability and resolve the information asymmetry problem.They can be applied in different situations. Then,in order to regulate the behaviors of selfish nodes,the dissertation proposes a reputation-based market incentive model.Different service levels can be traded for different prices.Service consumers buy service according to the system’s recommendation derived from reputation mechanism.Service providers are modeled as learning agents.They learn from the environment and arrive at optimal dynamic service level policies that optimize their benefit in the long run.The proposed model can simultaneously guarantee the interests of both the providers and consumers,thus, it can stimulate the selfish nodes actively contribute and consume resources.Afterwards,the dissertation proposes a reputation model in P2P file-sharing system based on punishment.Aiming at the disadvantages of the recent reputation models,enlightened by economic theory,the dissertation introduces the concept of punishment to the research on reputation mechanism and gives a punishment-based reputation model.The model can restrict non-collaborative behavior efficiently and stimulate sharing.Finally,the dissertation proposes a scorc and monitoring based incentive mechanism in P2P media streaming system.Each peer obtains a contribution score in two ways:uploading data to others by itself and by its child peers.The peers with higher scores have better chances to download data,To maximize the score,a peer can choose to monitor its child peers to stimulate sharing.The proposed incentive mechanism can solve the long-term history score problem,increase the cooperative proportion and maintain higher system performance.From the present problems in open networks that the characteristics of network dynamicity,heterogeneity between peers and unreliable resources,this dissertation deeply studies the unreliable service usability and Free-riding problems,proposes reputation-based resource transaction mechanisms and a market incentive model, provides a reputation model based on punishment in P2P file-sharing system and a score and monitoring based incentive mechanism in P2P media streaming system.It gives the brand new view and fine perspective for solving the problem of open network resource sharing issues.更多还原