《图数据库Neo4j架构实践》.ppt

图数据库 Neo4j 架构实践 技术创新，变革未来 关于此分享 内容 受众 “…使用图结构(graph structure)来实现语义查询，使用 点(vertex)、边(edge)和属性(property)来表示和存 储数据。最重要的概念是边(关系)，在存储中，它直接将 数据项关联起来…” /wiki/Graph_database 图数据库 图数据库 Social Network Facebook / Linkedin 场景 Spatial/GIS 路径规划 交通预测 场景 Recommendation “You may also like…” “People you may know …” 场景 Fraud detection First-party bank fraud/Insurance fraud/e- Commerce fraud 场景 Workflow engine Access control … 场景 场景 场景 Graph Query Response Time = ƒ(graph density, graph size, query degree) graph density: avg. (relations per node) graph size: ∑(node) query degree: hops per query 场景 场景 LinkedIn Flagship(职场人士) LinkedIn Students(学生群体) LinkedIn Influencer(职场影晌力内容发布和传播) LinkedIn Job Search LinkedIn Recruiter(企业征才解决方案) LinkedIn Lookup(企业雇员信息解决方案) Lynda(全球最大在线职业教育培训) Slideshare(全球最大的在线演示文档共享) Pulse(阅读和内容聚合) 自景 自景 选型 选型 OrientDB 缺少批量写，从DB文件加载并初始化非常慢 在200M+节点后写瓶颈 不健壮不稳定 CEO不靠谱 / 2015/06/the-orientdb-issues-that-made-us-give- up.html 选型 选型 Titan 更像是API layer，可选backend engine， Cassandra、HBase、BDB和ElasticSearch ACID不完备，依赖backend engine Key index使用限制较多 缺乏客户案例 Dev freeze The last man standing, and the winner is 选型 Neo4j Schema-free with labeled property Fast graph traversal(~2M/s) Written in Java Full ACID compliant Full/incremental online backup HTTP API or embedded in JVM Powerful declarative query language Cache(chunk) based sharding Linearly scalable read, M-S replication, fault-tolerance Plugins & extensions … Neo4j Neo4j Distributed replication fault tolerance scalability consistency Neo4j Cluster Management member join/leave/heartbeat Failover election status propagation Replication distributed lock transaction pull/push Neo4j Multi-Paxos algorithm less complex and testable Ready for cluster Management and election Neo4j • Multi-Paxos algorithm • less complex and testable • Ready for cluster Management and election Neo4j • Multi-Paxos algorithm • less complex and testable • Ready for cluster Management and election Neo4j • Multi-Paxos algorithm • less complex and testable