《基于dht的p2p网络资源存储查打技术研究》由会员分享，可在线阅读，更多相关《基于dht的p2p网络资源存储查打技术研究（65页珍藏版）》请在金锄头文库上搜索。
1、 I II 摘 要 I 摘要 目前，互联网系统的模式正在发生从传统的客户机/服务器（client/server） 模式到对等计算（peer-to-peer，亦简称 P2P ）模式的转变。P2P 的核心思想是所 有参与系统的节点（指互联网上的计算机）处于完全对等的地位，没有客户机和 服务器之分，也可以说每个节点既是客户机，也是服务器；既向别人提供服务， 也享受来自别人的服务。实际上，对等计算的概念在很早以前就已提出，但一直 没有受到广泛的重视，主要是因为没有实际运行的系统作为背景。产业界和研究 界都普遍认为在大多数情况下还是客户机/服务器模式更为合理。然而，随着 PC 技术和互联网（Intern
2、et）的发展，个人电脑的计算能力越来越强，接入带宽也逐 渐增大，如何更好地利用所有节点（尤其是原先处于服务器地位的节点）的能力 搭建更好的分布式系统自然而然地成为人们关注的问题。 由于系统中所有节点都是服务器，这样，将产生海量的数据。在如此海量的 数据中，如何快速有效地检索到目的资源，就成为 P2P 面临的首要问题。为了对 P2P 的信息进行查找，通常都在实际 P2P 网络之上抽象出来一个逻辑层，称作 Overlay 层，它把底层物理网络中的 P2P 节点提取出来，重新组成一个新的逻辑 层面上的网络。 以Napster为代表的基于中央日录服务器的P2P 系统和以Gnutella 为代表的非结构
3、化 P2P 系统不能提供有效且可扩展的检索机制。针对这个问题， 近年来提出了 CAN(Content-Addressable Network， 内容寻址网络)、 Pastry、 Tapestry 和 Chord 等采用 DHT(Distributed Hash Table，分布式哈希表)技术的结构化 P2P 系统， DHT 最基本的功能是提供资源的精确定位(Lookup)操作， 这是目前扩展性 最好的 P2P 路由方式之一。 然而现有的 DHT 系统在定位目标资源时，都存在寻路延时大、路由效率低 下的问题，这是由于这些系统忽略了节点在地理上的邻近性:一般根据节点标识 符 NodeID 来组织逻
4、辑层的重叠网络， 而标识符通过哈希节点的 IP 地址产生， 这 就使得节点在物理网络上的位置信息遭到破坏， 重叠网络上的逻辑相邻节点在实 际物理网络上可能相距甚远；另外不同节点的资源包括带宽、CPU 处理能力、 存储空间等存在着很大的差异， 然而现有的大部分 DHT 系统并没有考虑这一点， 而是将系统中所有节点都视为完全等同的。 这样势必导致查找过程的实际时延较 大，寻路效率低下。 基于 DHT 的 P2P 网络资源存储查找技术研究 II 为了解决问题，本文提出下列方法： 1、利用组来匹配物理网络，利用超级节点缓存查询和地址信息，并且在普 通节点收到查询请求时缓存查询源节点地址信息的路由算法
5、GcChord(Group Cache based Chord)。仿真结果表明，GcChord 的路由性能优于 Chord 系统，并且 节点重复(有一定倾向)查询越多，其平均物理、逻辑跳数越少。 2、在分析无结构与有结构 P2P 网络结构的基础上，文章提出一种基于节点 异构性的混合自适应 p2p 网络，网络分为 DHT 层和普通层。节点加入网络时并 不直接加入 DHT 层，而是选择一个 DHT 层节点作为代理，发布、撤销、搜索信 息。只有节点在线时间达到一定阈值或者其 DHT 代理的能力达到上限时，节点 才真正加入到 DHT 层。DHT 层节点在接近能力上限时自动转移负载。由此保证 DHT 层
6、节点都是较稳定、能力较强、状态良好的节点。从而减少 DHT 的路由维 护开销，提高查询效率。仿真结果表明，网络的稳定性和查询效率都有了明显的 提高。 关键字：P2P、Chord、组、拓扑匹配、负载均衡 Abstract III Abstract Nowadays, Internet system is transforming from the traditional mode of client / server into peer-to-peer (P2P for short) system. The core of P2P is that all participant computer
7、 peers (certain computer access to internet ) are equal to each other with no difference between client and server. That is, every peer is both client and server, therefore it offers service to other computers and meanwhile share service with others. The concept of P2P was brought forward earlier, b
8、ut people have paid little attention to it, mainly because there is no practical system as background. Its pervasively accepted by industry and research scholars that such mode of client/server is more reasonable in most cases. With the development of PC skills and Internet, the computing power and
9、bandwidth of PC is enhanced gradually. Accordingly it attracts the most attention how to build a better distributed system by fully utilizing all peers (especially those peers which are former servers). In the P2P system, all the peers are servers, so there are tremendous data in the Whole System. A
10、 fundamenta1 issue that confronts a large-scale P2P system is the efficient localization of the node that stores the desired date item. However, the first generation of P2P systems did not solve the problem well. Napster adopts a centralized catalog server but scalability is limited by the machine p
11、ower and the network bandwidth of the central point.Gnutella employs a messaging mechanism based on flooding, which can impose heavy burden on networks and thus compromise lts scalability. To address the problem, several research groups proposed DHT (Distrlbuted Hash Table) systems independently, wh
12、ich include Chord, CAN, Pastry and Tapestry. The fundamental function of DHT is accurate localization of the resource which has the best expansibility curently. However, DHTs are still faced with many problems such as great latency and inefficiency in routing. This is becanse DHTs neglect the neighb
13、ourhood and the resource differences among the nodes. DHTs reorganize peers into an overlay in the application level and route queens through the overlay using the node identifiers which are the hash of IP. This will destroy the neighbouthood among nodes. On the 基于 DHT 的 P2P 网络资源存储查找技术研究 IV other si
14、de, the nodes accessed to Internet are differentiated in the capacity of CPU and network bandwidth which will lead to greatd diference in relaying the messages. Its because that both of the factors are result in great latency and inefficiency in routing. To address the problem.we propose methods as
15、follow: 1) Give an idea that a topology-aware routing approach which uses groups and caches-GcChord(Group Cache based Chord). Experiment results demonstrate that GcChord improves the routing performance effectively, especially when nodes are apt to send some query more than others, both of the avera
16、ge logical hops and physical hop are improved observably. 2) P2P networks can be classified into unstructured and structured network. An unstructured P2P network is simpler but has high searching cost .While structured P2P networks may not be suitable for a highly dynamic environment .This paper shows the idea that an Adaptive Hybrid P2P Networks Based on Node Diversity which is divided into two layers-DHT layer and common layer. Node first join into common layer and choose a DHT layer n
