节点文献

中国教育科研网格数据管理中分布式副本定位模型的研究

Research on Distributed Replica Location Model in ChinaGrid Data Management

【作者】 王成伟

【导师】 金海;

【作者基本信息】 华中科技大学 , 计算机系统结构, 2006, 硕士

【摘要】 网格作为一项新兴的分布式高性能计算技术正在人类生活中发挥着越来越重要的作用。网格中的数据管理是其关键技术之一,它需要一个高效的副本定位模型来解决网格所处的广域网环境下的副本定位问题。中国教育科研网格(ChinaGrid)公共支撑平台(CGSP)是一个支持多个网格应用的通用性平台,在这个平台,数据按应用的不同分布在不同的逻辑域中。在CGSP的开发过程中,发现网格数据管理中的副本定位请求有两个特点,一是在绝大多数情况下,由于不同逻辑域间的安全因素,副本的映射会存放在它的数据文件所属的逻辑域中;二是在绝大多数情况下,用户对副本定位的请求在副本映射所产生的逻辑域发起。因此如何利用这两个特点,保证副本定位的安全性和效率成为了一大挑战。目前提出的众多副本定位解决方案都很难满足一个多应用逻辑域共存的网格环境下的副本定位需求。在新型的分布式副本定位模型,RSS(Replica Service System)中,副本服务点按逻辑域内和逻辑域间组织成一个分布式的多重环拓扑结构。在RSS模型中,副本映射按其逻辑域属性分为全局副本映射GRM(Global Replica Mapping)和本地副本映射LRM(Local Replica Mapping)。副本的定位过程考虑副本的逻辑域属性返回相应的副本映射,以满足用户的副本定位需求。RSS具有本地性,自组织性和域内负载平衡的优点。Boundary-Chord是RSS的核心算法,作为一个分布式哈希表(DHT)算法,Boundary-Chord降低了副本定位时物理层和逻辑层路由跳数,保证了对象分布的可控性。通过仿真模拟试验,RSS与已有的分布式副本定位模型P-RLS(Peer-To-Peer Replica Location Service)相比,具备更高的副本定位性能。测试的结果也表明,Boundary-Chord与其他分布式哈希算法相比在副本定位方面具有一定的优势。

【Abstract】 The emerging grids need an efficient replica location model to solve the replica location problem.In the experience of developing the ChinaGrid Supporting Platform (CGSP), a grid middleware that addresses building a uniform platform supporting multiple grid-based applications, we found characteristics of locality in the process of replica location. One is that replica mappings are stored in the logical domains they belong to with high probability for security reasons. Another is that a query for replica mappings is initialized with high probability in the logical domain where the replica mappings are generated. Therefore, it has become the main challenge to build a replica location mechanism that can make use of these properties of locality to guarantee replica location performance and security. Some previous works have been done to build a replica location mechanism, but they are not suitable for replica location in a grid environment with multiple applications like ChinaGrid. In this paper, we present a distributed replica location model, Replica Service System (RSS). In the model, two kinds of replica mappings, Global Replica Mapping (GRM) and Local Replica Mapping (LRM) are defined based on domain properties. RSS can locate these replica mappings to reply to users’query about replica locations, and it has the merits of locality awareness, self-organization, and domain load balancing, Boundary-Chord is the key algorithm of RSS. It has the merits of statistically less message routing hops on both application-level and IP-level, and the manageability of data placement.In simulation experiments, RSS outperforms present distributed replica location model, P-RLS. Simulation results show that the algorithm has better performance than other structured DHT solutions to the replica location problem.

  • 【分类号】TP393.07
  • 【被引频次】1
  • 【下载频次】92
节点文献中: