As the data organizations manage and produce grows increasingly complex, non-relational databases have become a preferred solution for storing, querying, and analyzing a large volume of data. Sometimes referred to as NoSQL (short for Not Only SQL) databases, non-relational databases are a popular alternative to the more typical relational database models that use SQL for simple data queries in column- and row-style tables.
Non-relational databases are best for big data
Non-relational databases are used for “big data” needs and real-time applications that generate diverse types of data. This unstructured data includes images, videos, social media posts, and documents — formats that prove challenging for a classic SQL-based database. SQL queries require more structured columns of data or numbers. However, large volumes of data or different data formats (images, social media posts) don’t fit in these classic spreadsheet columns — requiring a more non-relational database approach. As big data has become more complex and varied, non-relational databases are on the rise as the preferred choice for organizations. But what are the non-relational databases out there and how can you choose the best one for your project?
Types of non-relational databases
Non-relational database solutions are usually classified within four categories. Data infrastructure teams should analyze which option will best serve their data schema rather than try to rebuild data systems around an inflexible solution. These categories include:
Document store
This is the most common of the four categories, and the data is represented in document collections, usually JSON documents. These documents are “filed” in the database and can contain any amount and type of data. There are no constraints for the data contained in these documents. Examples of document stores include MongoDB and Elasticsearch.
Column store
This category is used to store big data across several data sets. It has similarities to relational databases as its structure supports SQL-style queries, but the columnar structure makes it possible to create highly optimized tables for specific kinds of queries. Examples of column-oriented systems include Amazon Redshift and Google BigQuery.
Graph store
A graph store manages data about network systems, which can grow to several terabytes of data. They’re optimized for querying data that’s organized into nodes and edges (which connect nodes), such as connections between friends in a social network. These are highly specialized systems; examples include Oracle’s “Oracle Spatial and Graph” product and Amazon Neptune.
Key-value store
The key-value stores a key, which is the identifier for the value. Data is looked up using the key or the value. There are no constraints with the data because you simply have key-value pairs for each element. It’s ideal for large-scale applications with queries structured to quickly return specific information for a user, item, or entity. If the value retrieved is a document, then a key-value store may also be classified as a document store. If metadata is associated with the documents, it may also be described as an object store. Key-value examples include Redis, Dynamo, and MemcacheDB.
3 popular non-relational database management systems
Thanks to the global growth of unstructured data, plenty of non-relational database options are available from relatively newer vendors to legacy companies with decades of big data experience.
1. MongoDB
Launched in 2008 as the “next-generation database,” MongoDB is a document-oriented database model. Companies like Bosch, Coinbase, Barclays, and Infosys use MongoDB to manage large quantities of unstructured data. Objects within MongoDB are defined by the user and organized into any sort of hierarchy the user desires. If you aren’t sure of the scale and type of data your application will be using, choose MongoDB for its flexibility.
Pros: MongoDB offers an expressive object model allowing users to index structured data efficiently and secondary indexes to query it more flexibly.
Cons: Some downtime may be experienced in the event of a node failure.
2. Cassandra
Released in 2008 by Apache, Cassandra is a column store used by companies like Apple, Intuit, Microsoft, and Netflix. It uses CQL, a proprietary query language similar to SQL. Though it has similarities to relational databases, Cassandra offers greater flexibility and scalability.
Pros: Cassandra has high availability, meaning the data stored within it will remain available even if a failure, like a hardware malfunction, were to occur.
Cons: The column structure does not offer the flexibility of MongoDB’s object model.
3. HBase
Apache’s HBase is another example of a column store non-relational database. It runs on top of Hadoop and HDFS (Hadoop Distributed File System) and is often referred to as the Hadoop database. With HBase, you can query records while viewing analytics reports across massive data sets.
Pros: Tight integration with Hadoop clusters. Its columnar nature makes it a good fit for quickly serving requests of a known structure.
Cons: Some downtime may be experienced in the event of a node failure.
While non-relational databases have grown in popularity over the last decade to process data generated with smartphones and Internet of Things (IoT) devices, relational databases remain an important piece of data architecture, especially when processing sensitive customer data. In fact, relational and non-relational databases can work together within a company’s data infrastructure to offset the other’s shortcomings. Structured data will often be stored in a relational system for offline business intelligence analysis. Data needed to support real-time applications can be exported into a non-relational database as needed by a website or application.
Source: https://business.udemy.com/blog/why-non-relational-databases-are-on-the-rise
For another insight of which elearning you can choose, you can see at https://coursemarks.com/course/learn-sql-for-data-analysis-with-google-big-query/