基于列数据库和图缓存的海量RDF管理-计算机应用技术专业论文.docxVIP

万方数据 万方数据 摘要 摘 要 Linked Data 的发展带动了各领域以 RDF 格式开放领域知识库，由此带来了 如何高效组织和管理海量 RDF 数据的问题。目前，基于关系数据库或 NoSQL 图数据库的存储模型，都不同程度地暴露了查询效率的问题。综合以上两种主流 存储模型，本文提出了底层磁盘采用基于列的关系存储模式，并且在内存中构建 RDF 图模式的分布式存储方案。 论文首先列举目前常见的 RDF 存储方式，分析了各自的优缺点，并论述了 基于列数据库 MonetDB 和图模式缓存管理大规模 RDF 数据集的可行性，然后提 出了系统架构：在磁盘上，基于 MonetDB 采用列式划分算法存储 RDF 数据集， 同时，将部分 RDF 数据以图模式缓存在内存中；查询时，会优先查找内存，然 后查找数据库，减少磁盘访问，提高查询效率。主要实现了以下几点：（1）基于 列式数据库的 RDF 分布式存储；（2）内存中构建 RDF 图模型；（3）存储节点处 理 SPARQL 查询的机制，包括缓存查找和数据库查询；（4）合并最终的结果集， 并且提取出需要的数据；（5）通过开放数据集，验证了该模型在数据存储、查询 方面的性能优势。 关键词： 海量 RDF；列式数据库；图模型；缓存查找；SPARQL 查询 I Abstra Abstract The development of Linked Data has led some areas publishing domain knowledge in RDF format, which brought the issue of how to organize and manage massive RDF datasets efficiently. Currently, the storage solutions based on Relational Database or Graph Database of NoSQL, have shown query performance problems. Based on the above two mainstream storage model, this paper presents distributed storage solution based on column-oriented Relational Database on disk, and construct RDF graph in memory. Firstly, the paper list some common RDF storage, analyzes advantages and disadvantages of each, and discusses the feasibility of managing large-scale RDF datasets based on column-oriented Relational Databases MonetDB and graph cache. Then it proposed the architecture: using vertical partitioning to store RDF dataset based on MonetDB on disk, also, caching part of RDF data into memory based on graph models; while querying, memory will be first to search, following by database, which reducing access to disk and query performance will be promoted. The main achievement are following: (1) distributed RDF storage based on column-oriented database; (2) building graph model of RDF in memory; (3)the solution of storage node processing SPARQL queries, including cache lookup and database search; (4) merge query datasets and extract data needed; (5)validates the performance advantages in RDF datasets storag