【作者】 高伟；

【导师】 张国印；

【作者基本信息】 哈尔滨工程大学 ， 检测技术与自动化装置， 2011， 博士

【摘要】 下一代新型计算机网络体系结构不断兴起。作为重点发展的P2P网络体系结构及相关技术是目前的研究热点。在本质上P2P是一种逻辑意义上的覆盖网络。P2P强调对等计算思想,网络中的节点具有平等的地位、权利和义务,每个节点既可获取资源又需提供资源,并参与到网络管理之中。为保证节点的自治性,允许节点动态加入或退出网络。建立在Internet之上的P2P的分布式结构造成资源的无组织,为了检索所需资源,节点必将广播其需求,洪泛转发的结果虽可找到资源,但浪费带宽。构造合理的覆盖网络,尽可能减少路由跳数,可在分布式结构中尽快地到达目标节点,这样可以提高系统搜索效率。节点的自治性、匿名性导致节点个人利益最大化的滋生,产生了搭便车现象,服务不可靠,甚至是恶意的攻击等行为。建立一套合理的信任机制,可保证数据的可信性和安全性。如果信任数据存放在非固定、非稳定的节点之上,P2P的动态性致使数据容易丢失或付出过大的维护开销。建立合理的信任数据管理机制,能有效保存信任数据,并对之进行有效的管理以维护网络动态性,提升系统综合性能。为此,论文对搜索策略,信任机制进行深入研究具有理论意义和实际应用价值。论文针对下述几个方面展开工作。本文提出了一个由物理网络层、覆盖网络层、信任层和信任数据管理层组成的四层的覆盖网络结构POTM。提出了一种基于极坐标的数据搜索定位协议,将CAN协议的多维笛卡尔坐标空间映射到一个极坐标空间中,并采用幂级数算法进行子空间的扩展,完成跳跃式查询,大大减少了逻辑路由跳数,提高了资源定位效率。采用证据理论进行信任建模,可融合多方信息,但无法有效聚合冲突。本文提出了一种新的证据合成规则NECR,引入权重系数,并对冲突证据进行处理,可提高证据合成的可靠性与合理性。论文基于NECR规则建立了一种新的信任模型,通过比例向量描述对命题的支持度,使得信任度的计算更准确灵活,可提高数据的可信性。主观逻辑理论从事实空间和观念空间两个不同的角度描述和度量信任关系,但没有考虑时间和否定事件对节点信任值的影响,也无法消除恶意行为的影响。本文提出了一种基于主观逻辑的信任模型NTMBSL,引入遗忘因子和采用变异系数描述否定事件评估风险因子计算信任值。将节点按功能分成群组,把节点之间的信任关系延伸到群组、节点之间的多方关系,增加信任度量的准确性,有效识别恶意节点,并可防止节点之间的协同作弊,从而提高整体安全性。为了对信任值进行综合管理,提出了信任数据管理模型,采用非劣解的管理者选择策略NISMS和基于差分指数模型的预期在线时间估计方法,基于综合权值对节点按组进行管理。设计了高效算法完成主节点、辅节点的选择过程,以及交换组间、组内消息,并设计相关算法规范了节点加入和离开行为。以上对信任数据的管理机制,可在一定程度上解决瓶颈问题,均衡负载,减少网络流量,提升系统整体可用性。更多还原

【Abstract】 A new generation architectures of computer network continues to burgeoning. As a key development in the P2P network architecture, its related technologies are current hotspot. In essence, P2P is an overlay network from the perspective of logical sense. P2P stressed the idea of Peer-to-Peer Computing. In P2P, nodes are equal in status, their rights and obligations are of the same. Each node can access resources, while they also need to provide resources. They are also involved into the network management. To ensure the autonomy of nodes, P2P allows nodes to dynamically join the network, or exit from the network.As P2P which basis is Internet, has a distributed structure, it makes that the resources are not organized. In order to retrieve the necessary resources, nodes must broadcast their requests. Because of using the flooding, it may finally find the resources, but a waste of bandwidth happens. If we can construct a reasonable overlay network, and minimize the routing hops in the network with a distributed architecture, we can quickly reach the target node. Thus, it can improve the search efficiency. Since nodes have the nature of anonymity and autonomy, which led to the fact that the pursuit of the node is to maximize their own interests. In the network, it generated some free-rider phenomenon. Therefore, service becomes unreliable, and even in some cases it may appear the behavior of malicious attacks. Our plan is to establish a reasonable mechanism for the trust, in order to ensure the credibility and security of data. Trust data, they need to be stored in non-fixed and non-stable nodes. The dynamic nature of P2P leads to the fact that the data likely to be lost, or it’s too costly to maintain. For the trust data, we build a reasonable management mechanism in order to effectively preserve the trust data, and effectively manage them, so that it can maintain the dynamic nature of the network and enhance the performance of the system. To this end, we focus on the search strategies and the trust mechanisms, because of their theoretical and practical value.In this thesis, we started our research work on several aspects mentioned earlier.Existing retrieval systems can not fully support the resource organization, the Resource Location is inefficient. This thesis presents a four-layer overlay network structure, we call it "POTM", which consists of the physical network layer, the overlay network layer, trust layer, and management layer of the trust data. This thesis proposes a protocol of data search and resource location, which based on polar coordinates. The multi-dimensional space of CAN protocol is re-mapped to another polar coordinate space. We combines Resource space model with the unified organization of distributed data resources, thus it constitutes an overlay network layer. In this thesis, we use the power series algorithm which extended the subspace, thus we greatly reduces the logical routing hops, and improve the efficiency of data search.Based on evidence theory, we model the trust mechanism. It can integrate multi-information, but can not effectively aggregate conflict. This thesis proposes a new rule of evidence combination, we call it "NECR". We introduce the weight coefficient, and deal with the conflict of evidence, in order to improve the reliability and reasonable of evidence combination. Based on NECR rule, we have created a new trust model. By comparing the degree of support for the proposition of proportion vector, we can get a more accurate and flexible calculation for the trust data, and improve the credibility of the data.The theory of Subjective logic descripts and measures the trust relation from the two views of real space and concept space. However, it did not take into account of impact of time and the negative events for the nodes’trust value, and it also can not eliminate the impact of malicious behavior. This thesis presents a trust model based on subjective logic, we call it NTMBSL. We introduce the forgetting factor and the risk factors, in order to calculate the trust value. The nodes are divided into groups by function. We extend the trust relationship between nodes to the group. Multiple relationships between nodes can increase the confidence in the accuracy of measurement, effectively identify the malicious node, and it prevents cheating collaboration between nodes to improve overall security.We propose a trust model for data management, which manage data by group, so it can complete the integrated management of trust value. We design the corresponding algorithm, it selects the master node, secondary node, and exchange messages within the group or groups, and it provides the nodes’behavior of join and leave. The algorithm can solve the bottleneck problem to some extent, balance the load and reduce network traffic, and improve overall system availability.更多还原