动态可重构的DSM语义研究

发布时间：2018-01-16 03:04

本文关键词：动态可重构的DSM语义研究　出处：《哈尔滨工业大学》2012年硕士论文　论文类型：学位论文

【摘要】：在信息大爆炸的时代，分布式系统是现今计算科学的主要研究重点之一。随着用户及信息数据的爆炸式增长，系统在发生着剧烈的变化，新的问题也显现出来。各节点各自的行为、链路失效、节点故障和其他的环境因素影响了各独立节点的可用性，系统的动态性趋势越来越强，如何在动态的环境下对系统进行重构，，已经成为重要的研究方向。本文在此背景下，对动态环境下的分布式存储系统如何进行重构并实现进行了研究。首先，对分布式共享存储系统进行研究，通过对传统共享技术的比较，确定了满足重构要求而需要的一致性保证，动态性机制，系统模型等要素并加以分析和研究。通过深入分析，提出了动态重构要达到的目标。针对分布式存储系统节点故障，链路失效，节点动态的加入和删除等问题，本文在深入的研究了各种重构算法的基础上，提出了FDRDSM的快速重构算法。首先是对算法的研究和实现，包括读写，重构及垃圾回收的实现，通过这几个方面的实现达到对分布式存储系统重构的目的。其次，通过对算法的一致性分析，保证了算法的原子性，确保操作能按顺序执行。最后通过在基于Lustre分布式文件系统的环境下实现了该算法，通过对各服务器的动态重构及数据分析，验证了算法的可行性。
[Abstract]:In the era of information explosion, distributed system is one of the main research focus of computer science. With the explosive growth of users and information data, the system is undergoing drastic changes. New problems also appear. The behavior of each node, link failure, node failure and other environmental factors affect the availability of independent nodes, and the dynamic trend of the system is becoming stronger and stronger. How to reconfigure the system in dynamic environment has become an important research direction. Under this background, this paper studies how to reconstruct and implement distributed storage system in dynamic environment. First of all, the distributed shared storage system is studied. Through the comparison of the traditional sharing technology, the consistency guarantee and dynamic mechanism to meet the requirements of reconfiguration are determined. The system model and other elements are analyzed and studied. Through in-depth analysis, the goal of dynamic reconstruction is put forward. Aiming at the problems of node failure, link failure, node dynamic join and delete in distributed storage system, this paper studies various reconstruction algorithms deeply. First, the research and implementation of the algorithm, including the realization of reading and writing, refactoring and garbage collection, is proposed. Through the realization of these aspects to achieve the purpose of reconfiguration of distributed storage system. Secondly, through the consistency analysis of the algorithm, the atomicity of the algorithm is ensured. Finally, the algorithm is implemented in the environment of distributed file system based on Lustre, and the feasibility of the algorithm is verified by the dynamic reconfiguration and data analysis of each server.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP333

【参考文献】