Graph Database Comparison and Analysis (Storage&Clustering)

Big Data Era has become increasingly important for discovering the values by managing data, not simply read and wrote that. As a way of this, Graph Database that storing and analyzing the relationship between data is receiving attention. In computing, a graph database is a database that uses graph structures for semantic queries with nodes, edges and properties to represent and store data. It also can be used for the social network, recommendation system for the shopping mall, fraud detection, and so on.

TOP 3 of the present graph databases are Neo4j, OrientDB, and Titan. So, we are going to tell you how they store and process the graph model.

According to the above figure, Neo4j is currently the leading database, and has become the market leader. Neo4j is the Labeled property graph model, and it has the vertex and edge, which offers the property. It can also represent as grouping the vertex and edge to label. Neo4j is able to have the label and add or remove that.

Neo Technology, made by Neo4j, emphasizes the possession of native storage and process for graph when Neo4j promotes its products. So, it was specially designed for specializing a storage and processor to the graph model unlike the competitors. Especially it did the direct pointing through pointer(id) of the record(related object) unlike RDB which processes data by JOIN. It hereby enables the graph traversal to guarantee high speed between the neighboring nodes.

A way of implementing Direct pointing is that loads the each element(node, relationship, and property) in the separate store file. The each element allocates ID of the loading position and is implemented for a way of finding the record’s location by record offset(ID * record size).

OrientDB, a multi-model database, has the merit of constructing Cluster combined several data models by itself because it provides the document, graph, key/value, object model. However, bucket of key/value and class of object model are stored on the object of class(like table of RDB), and pair and object of key/value are stored on the document(like record of RDB). So OrientDB offers key/value model by using the document model which provides many constraints.

OrientDB uses a concept called cluster when it store the record(vertex, edge, and document). Cluster is actually an object loading the records. Each class has one more cluster and Each cluster can set up in which node(machine) is loaded. Class and cluster can set up sharding and replication how they define. When it defines the class, it defines the several clusters. If it sets up in which each cluster has some kind of data, it will be the sharding and if each cluster is duplicated in several nodes, it will be the multi-master replication. Provided, as the below figure, it supports only data sharding of class level. In other words, it does sharding that is partitioning class to cluster.

Titan is also a Labeled property graph model and is a database based on Java suitable for Big Data. Graph model have the schema of vertex label, edge label and property key. Edge label and property key can make schema by setting it definitively before data is entered, but can’t change type of schema, once done.

Titan provides the multiplicity for edge label and the cardinality for property key with no competitors. Also the best strength that Titan is better than them is to retain the storage backend and index back end of module format. It can choose and use Hbase, Cassandra, BerkeleyDB to storage and if the storage stores Big Table model, it can be added to Titan’s storage due to offer storage of module form. The storage structure of Titan consists of vertex centric. Graph is the collection of vertexes and each vertex has the adjacency list. Like below, this storage has an advantage that can find related edge and property in one approach.