Comparison of Graph Database: Neo4j, JanusGrraph and HugeGraph

最新推荐文章于 2024-09-28 14:37:43 发布

Jermy Li

最新推荐文章于 2024-09-28 14:37:43 发布

阅读量1.1k

点赞数

分类专栏：图数据库文章标签：数据库知识图谱 database

本文链接：https://blog.csdn.net/javeme/article/details/122312685

版权

HugeGraph 同时被 2 个专栏收录

26 篇文章 38 订阅

订阅专栏

图数据库

24 篇文章 24 订阅

订阅专栏

                    
                        
                    
                    Main FeaturesNeo4jJanusGraphHugeGraph
Open Source EcologyThe community version is open source, the commercial version is closed sourceOpen source, compatible with the Apache Tinkerpop ecosystem, mainly provided by IBM on cloud services.(Janus has added few features since 2015 forked from Titan)Open source, compatible with the Apache Tinkerpop ecosystem. (HugeGraph has continued to add a large number of features with active state since 2017)
Technology ArchitectureStand-alone version architecture, graph storage structure adopts adjacency linked list, suitable for scenarios where small-scale graphs can be accommodated in memory (linked lists are not suitable for query on disk)Share-Storage architecture, graph storage structure adopts adjacency sequence table, mainly adopts HBase as back-end storageShare-Storage architecture, Share-Nothing architecture (RocksDB), graph storage structure adopts adjacent sequence table, single-machine can support billion-level graphs, read and write performance is much higher than Titan/Janus
Data ScaleCommunity version billion-level, stand-alone versionOver 10 billionsOver 100 billion
Write PerformanceOnline import speed is slow (~10k/s) , Offline import speed is faster (10~100k/s)Slow (1~10k/s), especially for graphs above one billion level.Online import is fast (100~500k/s), and supports fast overwriting feature
Read Performance10~40 k/s，on 100 million scale graph~10k/s，with performance jitter20~100k, the performance of HugeGraph 0.12 is faster than Neo4j 2x+ on 100 million scales; HugeGraph is faster than Neo4j 5x on the 1 billion scale graph
Super VertexAdjacent edges query of super vertex is slow, and the cross-linked list storage structure is difficult to speed up the query partial of adjacent edgesCan be relieved by Vertex-Centric indexCan be relieved by Vertex-Centric index, and supports access all data by paging
Built-in Common Graph AlgorithmsProvides an installation algorithm package, providing graph algorithms like path search, similarity, centrality, community detection, link prediction, etc.Not SupportedBuilt-in provides basic graph algorithms, like path search, collaborative recommendation, centrality, community discovery, etc.
Support large-scale Graph ComputingNot SupportedSupport the expansion of Spark GraphX, Giraph, etc.Built-in HugeGraph-Computer, providing large-scale parallel graph computing, in addition to supporting the expansion of Spark GraphX
HASupported by Commercial VersionNot SupportedSupported

                

Main Features	Neo4j	JanusGraph	HugeGraph
Open Source Ecology	The community version is open source, the commercial version is closed source	Open source, compatible with the Apache Tinkerpop ecosystem, mainly provided by IBM on cloud services.(Janus has added few features since 2015 forked from Titan)	Open source, compatible with the Apache Tinkerpop ecosystem. (HugeGraph has continued to add a large number of features with active state since 2017)
Technology Architecture	Stand-alone version architecture, graph storage structure adopts adjacency linked list, suitable for scenarios where small-scale graphs can be accommodated in memory (linked lists are not suitable for query on disk)	Share-Storage architecture, graph storage structure adopts adjacency sequence table, mainly adopts HBase as back-end storage	Share-Storage architecture, Share-Nothing architecture (RocksDB), graph storage structure adopts adjacent sequence table, single-machine can support billion-level graphs, read and write performance is much higher than Titan/Janus
Data Scale	Community version billion-level, stand-alone version	Over 10 billions	Over 100 billion
Write Performance	Online import speed is slow (~10k/s) , Offline import speed is faster (10~100k/s)	Slow (1~10k/s), especially for graphs above one billion level.	Online import is fast (100~500k/s), and supports fast overwriting feature
Read Performance	10~40 k/s，on 100 million scale graph	~10k/s，with performance jitter	20~100k, the performance of HugeGraph 0.12 is faster than Neo4j 2x+ on 100 million scales; HugeGraph is faster than Neo4j 5x on the 1 billion scale graph
Super Vertex	Adjacent edges query of super vertex is slow, and the cross-linked list storage structure is difficult to speed up the query partial of adjacent edges	Can be relieved by Vertex-Centric index	Can be relieved by Vertex-Centric index, and supports access all data by paging
Built-in Common Graph Algorithms	Provides an installation algorithm package, providing graph algorithms like path search, similarity, centrality, community detection, link prediction, etc.	Not Supported	Built-in provides basic graph algorithms, like path search, collaborative recommendation, centrality, community discovery, etc.
Support large-scale Graph Computing	Not Supported	Support the expansion of Spark GraphX, Giraph, etc.	Built-in HugeGraph-Computer, providing large-scale parallel graph computing, in addition to supporting the expansion of Spark GraphX
HA	Supported by Commercial Version	Not Supported	Supported