Kinds of database

Currently Databases are classified as either Relational or NoSQL

NoSQL itself is just a capture term to describe, none relational data structures.

As with RDBs having multiple flavours and brands

Relational Databases

• MySQL
• MariaDB
• MS SQL
• Oracle
• IBM DB2
• PostGres

NoSQL

• MongoDB
• CouchDB
• CouchBase
• Riak
• Cassandra

CAP Theorem

CAP Theorem, also knows as Brewer's Theorem from Eric Brewer

• Consistency: all nodes belonging to system see
the same data at the same time
• Availability: a guarantee that every request
receives a response of success or failure
• Partition tolerance: system continues to operate
despite message loss or failure of parts of system

Given a data partition, will you prioritize consistency or availability?

And this applies to Relation how?

When to use Relational Database Management Systems (RDBMS)

• Data normalization is critical for elimination of redundancy and ensuring master data consistency
• Many “justifications” for using relational databases are cultural, not technical e.g. Solution already built with relational DB  resistance to change
• Prioritization of data availability and consistency lends RDBMS well to handling transactional, reporting, log, and warehouse data
• Analytics and BI tooling are valid reasons for maintaining a relational database, but this too is rapidly changing

According to the CAP Theorem, it is impossible for relational databases to be partition tolerant

• So long as RDBMS prioritizes availability and consistency, they are
unable to scale out (horizontally)!
• Vertical scaling is the alternative, but this practice becomes prohibitively expensive and is not sustainable

And NoSQL fits in here…?

Most NoSQL technologies have been built for scale, which means that they fit either a CP or AP model

Popular datastores like CouchBase & Mongo try to be CP, but fallback to AP when things get tough

CouchDB follows an AP approach from the start using an eventual consistency

Key Value Stores

• Cassandra
• Riak
• MemcacheDB
• HBase
• pickleDB

Columnar Stores

• Cassandra
• HBase

Document Stores - 50% of NoSQL DBs are document based

• CouchDB and because of it Cloudant
• MongoDB
• Redis
• CouchBase
• Engine Yard

Graph Store

• *dex
• Neo4j
• InfiniteGraph
• Sesame

                
        {
             "firstName“ : "John",
             "lastName" : "Smith",
             "age“ : 25,
             "address" :
             {
                 "streetAddress“ : "21 2nd Street",
                 "city" : "New York",
                 "state“ : "NY",
                 "postalCode" : "10021"
             },
             "phoneNumber":
             [
                 {
                   "type" : "home",
                   "number“ : "212 555-1234"
                 },
                 {
                   "type" : "fax",
                   "number“ : "646 555-4567"
                 }
             ]
         }
                
              

And they all use JSON or some derivative, that's basically JSON but a different name

How does this apply to CAP theorem

As I said, all these datastore use or fallback to an AP approach. Which means…

Instead of prioritizing consistency and availability, shift focus towards ensuring availability and partition tolerance

• Unlikely to find a scenario where loss in availability would be tolerable
• Selecting for partition tolerance opens up possibility for horizontal scaling!
• Distribution over cluster also improves availability In aggregate the cluster is more reliable than the individual nodes that comprise it

This comes at the cost of a weakened consistency model

• Can no longer guarantee that all nodes (and clients connected to these nodes) share identical versions of the same data at a given moment
• The result is an “eventual consistency” model The premise that all nodes in a distributed system will eventually share the same versioning of all data, given sufficient time