P2P存储中辅助系统及数据有效性的研究

发布时间：2018-04-27 16:20

本文选题：P2P + 分布式存储　；参考：《南京大学》2013年硕士论文

【摘要】：随着网络技术的发展,带宽的提高,互联网应用发生了巨大的变化。存储在互联网上的数据越来越丰富,用户访问量也越来越大。这使得传统单一服务器提供存储的模式不再适应当前的形势。基于P2P的分布式存储由于其自组织、支持海量用户等特点,成为最适应当前数据环境下的存储结构,引起了人们的广泛研究。由于P2P存储中用户行为不可预测等原因,当前还没有商业化的P2P存储系统。一些学者开始探索将P2P存储做为辅助系统应用。随着云计算的兴起,云存储服务日趋成熟,许多公司开始向公众提供云存储服务。存在一些云存储在提供高效可靠的服务的同时,对用户从云中读取数据时占用的带宽也收取费用。已经取得广泛研究的P2P存储可以用作云存储的辅助系统从而有效的移除从云端读取数据所产生的费用。当前节点辅助系统都是基于无结构覆盖的。在P2P存储领域,存在众多的基于结构化覆盖构建的P2P存储系统。本文中,我们探索如何将基于结构覆盖的P2P存储作为辅助系统应用到云存储中去。通过在Chord组成的环形结构上顺序的存储数据,我们将读取数据协议的复杂度由以前的O(KlogN)降到了O(logN+K)。对任意类型的节点辅助的云存储系统,由于系统中的存储由云存储及P2P存储两部分构成,因此需要确定应由哪一方响应用户读取文件的请求。在基于结构覆盖的P2P存储中,由于不存在tracker收集系统的实时状态信息,因此为带宽分配方案的设计带来了困难。本文中我们提出了一种预测的方法来有效解决带宽分配问题。节点辅助系统设计的优势在于数据的有效性由云端负责,因此在P2P存储中不需要关注数据有效性的问题。但是在云存储的设计中,必须将数据有效性这一点纳入考虑。当前存在为数众多的工作研究如何使用P2P技术构建云存储系统,这些系统一般是由大量廉价的存储节点构成的封闭式P2P存储系统。由于软件错误,硬件失效等原因,会导致节点失效,使得存储在相应节点上的数据失效,因此需要设计相应的机制保证数据的有效性。本文中,我们提出了一种基于分组与备份的策略来保证数据的有效性。使用时间片模型,在假设节点有效状态可预知的前下,我们提出了最大分组与最优分组问题。并证明这两个问题的复杂度都是NP-Complete。在这之后,根据实际情况,加入节点在连续时间片上保持有效这一条件,我们设计了多项式时间的算法求解最大分组数这一问题。最后,通过将节点组织成有向无环图,我们使用PSA算法来穷举求解最优分组问题。
[Abstract]:With the development of network technology and the improvement of bandwidth, the application of Internet has changed greatly. More and more data is stored on the Internet, and more and more users visit it. This makes the traditional single server storage mode no longer adapt to the current situation. The distributed storage based on P2P has become the most suitable storage structure for the current data environment because of its characteristics of self-organization and mass user support. Due to the unpredictable behavior of users in P2P storage, there is no commercial P2P storage system. Some scholars began to explore P2P storage as an auxiliary system application. With the rise of cloud computing, cloud storage services become more and more mature, and many companies begin to provide cloud storage services to the public. There are some cloud storage to provide efficient and reliable services, while charging users the bandwidth they use to read data from the cloud. P2P storage, which has been widely studied, can be used as an auxiliary system for cloud storage to effectively remove the cost of reading data from the cloud. Current Node Auxiliary Systems are based on unstructured coverage. In the field of P2P storage, there are many P2P storage systems based on structured overlay. In this paper, we explore how to apply P2P storage based on structure coverage to cloud storage as an auxiliary system. By storing data sequentially on the ring structure of Chord, we reduce the complexity of the reading data protocol from the previous OKlogN to the O(logN KG. For any type of node-assisted cloud storage system, since the storage in the system consists of cloud storage and P2P storage, it is necessary to determine which party should respond to the user's request to read files. In P2P storage based on structure coverage, it is difficult to design bandwidth allocation scheme because there is no real-time state information of tracker collection system. In this paper, we propose a predictive method to solve the bandwidth allocation problem. The advantage of node-aided system design is that the validity of data is in the cloud, so there is no need to pay attention to the problem of data validity in P2P storage. However, in the design of cloud storage, the validity of data must be taken into account. There are many researches on how to use P2P technology to construct cloud storage system. These systems are generally composed of a large number of cheap storage nodes in a closed P2P storage system. Due to the failure of software and hardware, the data stored on the corresponding nodes will fail, so it is necessary to design the corresponding mechanism to ensure the validity of the data. In this paper, we propose a strategy based on grouping and backup to ensure the validity of data. Using the time slice model, we propose the maximum packet and optimal grouping problem under the assumption that the effective state of the node is predictable. It is proved that the complexity of these two problems is NP-Complete. After that, according to the actual situation, adding the condition that nodes remain efficient on the continuous time slice, we design a polynomial time algorithm to solve the problem of the maximum number of groups. Finally, by organizing nodes into directed acyclic graphs, we use PSA algorithm to solve the optimal grouping problem.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333

【相似文献】