基于调度的n元脉动通信模型的研究与实现

【作者】 陈小辉；

【导师】 刘心松；

【作者基本信息】 电子科技大学 ， 计算机系统结构， 2005， 硕士

【摘要】 据国际数据公司(IDC)资料显示:当前数据库高端应用正朝着“三高一大”(高性能、高可靠、高安全、大数据量)方向发展。分布式并行数据库管理系统恰好以其固有特性和设计目标,顺应了数据库应用的发展趋势;同时,应用领域也给分布式并行数据库管理系统的高性能和可靠性等提出了更高的要求。为实现分布式并行数据库管理系统的高性能,本文从通信子系统应合理适量使用系统资源出发,将动态调度思想引入到改善机间通信的研究中,提出了基于调度的通信模型。该模型根据系统运行情况,动态计算通信子系统应占用的最佳系统资源,并据此动态获取或释放系统资源,以期既克服分布式并行数据库管理系统通信子系统这一系统性能瓶颈,又不至于使分布式并行数据库管理系统的对外有效处理能力成为新的系统性能瓶颈,从而达到提高分布式并行数据库管理系统整体性能的目的。同时,为实现分布式并行数据库管理系统的高可靠性,及时检测出分布式并行数据库管理系统中的故障节点,实现系统故障自动重构功能。本文提出了n 元脉动通信模型。该模型将分布式技术与逻辑环有机融合,将分布式并行数据库管理系统中的节点组织成一个逻辑环,采用非阻塞TCP 完成分布式异步脉动过程,实现分布式并行故障检测机制,从而达到提高分布式并行数据库管理系统可靠性的目的。将基于调度的通信模型与n 元脉动通信模型融为一体,提出了基于调度的n元脉动通信模型。该模型已在分布式并行数据库管理系统DPSQL 中实现,为研究该模型对分布式并行数据库管理系统性能及其可靠性的贡献,通过编写了测试代码,对DPSQL 进行了对比测试和重负载测试;通过模拟故障,对故障节点检测进行了模拟测试。测试结果表明该模型对提高分布式并行数据库管理系统的整体性能和可靠性极为有效。本文主要分为四个部分:第一章分析了现有改善通信性能的各种技术及其相对于DPSQL 的局限性;第二章给出了分布式并行数据库管理系统及其特征,分析了分布式并行数据库管理系统服务器节点间通信的特殊要求;第三章详细论述了基于调度的n 元脉动通信模型及其实现;第四章简要分析了基于调度的n 元脉动通信模型的性能,描述了测试结果;最后进行了总结。更多还原

【Abstract】 According to IDC, the trend of application for database at high level goes ahead with demand of “high performance, high reliability, high security and huge storage”. Meanwhile, application field also needs database products to provide users with higher performance and higher reliability. By right of its inherent features and design goal, distributed and parallel database management system (DPDBMS) meets the trend of demand. To obtain high performance in the DPDBMS, this thesis, as viewed from that communication subsystem ought to properly make use of system resource, introduces dynamic scheduling strategy into the research of improving communication performance and brings forward new communication model based on scheduling. The model, according to current status of system, could dynamically calculate the most suitable system resource ratio of communication subsystem, then decides whether to get or free some system resource. In this way, not only performance bottleneck within communication subsystem could be overcome, but also couldn’t result in making external processing ability of DPDBMS become new bottleneck. As a result, high performance is gained successfully. To obtain high reliability, that is, DPDBMS could detect sites in failure status in time and provide auto recovery function while confronted with system failure. This thesis proposes n Dimension Pulsing Communication Model (nDPCM) which turns all sites in DPDMS into one ring by combining distributed method with logical ring, and adopts non-block TCP to finish distributed asynchronous pulsing process. In this way, high reliability is achieved successfully. Combining communication model based on scheduling with n Dimension Pulsing Communication Model, the n Dimension Pulsing Communication Model Based on Scheduling (nDPCMBS) is proposed and implemented in DPSQL (a kind of distributed and parallel database management system). Test result shows that nDPCMBS is an effective way in improving performance and reliability of DPSQL. This thesis contains four chapters as a whole. In chapter 1, various techniques of improving communication performance are reviewed. In chapter 2, DPDBMS, DPSQL and their features are discussed. In chapter 3, nDPCMBS and its implementation are discussed in detail. In chapter 4, both performance analysis on更多还原