Graphics Reference
In-Depth Information
Graph Databases
A system for dynamic interactive analysis of graphs begins with a data
storage and access solution. Relational databases like Oracle or SQL Server
have many points of strength but, contrary to the term itself, are not
particularly efficient for exploring the complex relations or relatively noisy
heterogeneous data typical of large graphs. Data within a table must
conform to rigid pre-defined schemas, and linking data across many tables
requires complex and expensive joins that limit query speed.
A significant amount of research has been invested in providing relational
database functionality on distributed systems. For example, Google Big
Query provides cloud-based data storage and query using a core subset of
SQL. Cloudera Impala is similarly a SQL query engine for Hadoop that
has progressed to the point where it may no longer be necessary to copy
subsets of data from the Hadoop Distributed File System (HDFS) to local
relationaldatabasesinordertoachieveinteractiveperformanceforanalysis.
The benefit of carrying SQL forward with these solutions is a well-known,
time-tested interface language that provides relatively straightforward
migration from traditional databases.
In parallel, work has been ongoing for alternatives to tabular storage and
SQL, known broadly as
NoSQL
or “Not only SQL.” Architectural
approaches include column-based, document-based, and key-value stores.
In addition, a number of graph databases such as Neo4J and Titan have
emerged, which represent data in a fundamentally different way. Instead of
storing and representing data as a regular set of dimensioned structures,
data is represented as a property graph where elements are nodes, and
relations between elements are represented as links.
Figure 14-2
shows a notional example of product data for two Blu-ray discs.
The products are associated by a single common categorization, but also
share reviewers. Both nodes and links in a property graph can include
properties. In this example, each review relation includes a rating from one
to five stars.
