NoSQL databases

NoSQL databases
Filip Ilievski (f.ilievski@vu.nl)
Vrije Universiteit Amsterdam

What will you hear about today
1) Databases: A historical perspective
2) Modern trends and why NoSQL?
3) Properties of NoSQL
4) Categories and instances of NoSQL databases
5) RDBMSs vs NoSQL
6) Use cases & The Semantic Web

The origin of Relational DBMSs
“In 1970, Edgar F. Codd, an Oxford-educated mathematician working at the IBM
San Jose Research Lab, published a paper showing how information stored in large
databases could be accessed without knowing how the information was structured
or where it resided in the database.”

“Until then, retrieving information required relatively sophisticated computer
knowledge, or even the services of specialists who knew how to write programs to
fetch specific information—a time-consuming and expensive task.”
“What Codd did was open the door to a new world of data independence. Users
wouldn’t have to be specialists, nor would they need to know where the information
was or how the computer retrieved it. They could now concentrate more on their
businesses and less on their computers.”
(http://www-03.ibm.com/ibm/history/ibm100/us/en/icons/reldb/)

After Codd’s paper described the relational ideas in theory, practical
implementations followed.
Soon, the Relational DBMSs were proven superior over the ad-hoc flat file
databases and became de facto standard in terms of modelling, representing and
accessing data.
Relational databases were (and often still are) ideal when the storage of data was
expensive and the data schemas were fairly simple.

Trend #1: Increase in modeling complexity
In the Big Data era, it is not uncommon for a database to have hundreds-
thousands of tables, and many of these should be combined to give the final
requested information.
=> Trend #1: Queries get more complex and require complex SQL
operations

Trend #2: Using tables is not always optimal
Think of social media, where it is more
intuitive to use something like a graph
network
=> Trend #2: Some use cases require
graphs or nested structures rather
than tables

Trend #3: A fixed schema is not always optimal
Our model might change over time
Applications in practice are typically incremental (“agile”), so the initial design
choices get altered -- and this influences the data model.
Having a fixed pre-defined schema is not always desired.

Trend #4: Increase in querying intensity
The Internet in the 80s was not there as we know it: only a handful of people
would access and use a database.
There are many more users accessing the data now, mainly needed by the social
networks and enabled by the computing power increase.

Trend #4: Increase in querying intensity
The Internet in the 80s was not there as we know it: only a handful of people
would access and use a database.
There are many more users accessing the data now, mainly needed by the social
networks and enabled by the computing power increase.
Additionally, the intensity of access by an individual user is higher (imagine that
each Facebook like is yet another WRITE to the database).
● A single website being accessed can trigger tens-hundreds of database
queries.
=> Trend #4: Many more queries have to be handled
=> Trend #4’: The querying intensity varies

Trend #5: Cheap storage
Hardware is much cheaper, more accessible and powerful now.

Trend #5: Cheap storage
Hardware is much cheaper, more accessible and powerful now.
When scaling up, it makes more sense to add more computers (horizontal scaling)
rather than to upgrade a single machine (vertical scaling)
=> Trend #5: Usage of a lot of cheap (cloud) hardware instead of a single
powerful machine

Modern trends: summary
=> Trend #1: Queries get more complex and require complex SQL
operations
=> Trend #2: Some use cases require graphs or nested structures rather
than table
=> Trend #3: The data model might evolve over time
=> Trend #4: Many more queries have to be handled
=> Trend #4’: The querying intensity varies
=> Trend #5: Usage of a lot of cheap (cloud) hardware instead of a single
powerful machine

Relational databases in the modern era
Relational databases are not designed to handle many users or large data.
Relational databases are not designed to be distributed over multiple computers.
Relational databases are not efficient with data that is not meant to be stored in
tables.
Sometimes, the relational database principles (e.g. normalization) are too complex
and slow when there is a lot of data or a complex model.

NoSQL (=Not-Only-SQL)
Main idea: adapt to the new trends/needs
1) To avoid making complex queries and joining many datasets, store as much
as possible in a single record
a) Consequently, the data structure in NoSQL is often not a table, but instead a dictionary, a tree,
an array, etc.
b) Yes, this leads to repetition and violates the normalization principle.

an array, etc.
2) Use a data structure that is most appropriate for the problem (e.g. use a
graph to model social networks)

an array, etc.
2) Use a data structure that is most appropriate for the problem (e.g. use a
graph to model social networks)
3) To handle many users and many queries, and to use the cheap hardware,
distribute the data over many simple machines.

What is NoSQL?
It is a group of databases that attempts to provide more flexible way of data
storage, while adapting to the new trends of intensive storage and accessible
hardware.
This idea started around 2009, and it has been already widely adopted.

What is NoSQL?
NoSQL stands for Not-Only-SQL (yes, some NoSQL databases support SQL
operations too).
NoSQL databases are NOT meant to replace relational databases, both have their
use cases.

● Flexible schema / schema less
Characteristics of a NoSQL database
Image from https://www.slideshare.net/mikecrabb/a-beginners-guide-to-nosql/30-SIDENOTEcolour_tabbyname_Gunthercolour_gingername_Mylocolour

● Non relational (forget about normalization today - but you have to know it for
the exam :) )

● Non relational
● Simple access compared to SQL (but not standard across products)

● Non relational
● Simple access compared to SQL (but not standard across products)
● Basically, does not support complicated queries
● Cheaper than RDBMS systems
● Horizontally scalable
● Replicated
● Distributed

NoSQL databases
Filip Ilievski (f.ilievski@vu.nl)
Vrije Universiteit Amsterdam
Half time

A)KEY-VALUE DBs
Each record in the database contains a unique key, which “hides” the value
Analogue to key-value pairs in dictionaries/JSON
Very simple and designed to work with a lot of data
Each record can have a different structure/type of data

A)KEY-VALUE DBs: Code for Redis
Initialization
import redis
pool = redis.ConnectionPool(host='localhost', port=6379, db=0)
r = redis.Redis(connection_pool=pool)
Store a record
my_key =”B”
my_value=”triangle”
r.set(my_key, my_value)
Obtain a record
my_key=”B”
r.get(my_key)

B) DOCUMENT-STORE DBs
Again, simple and designed to work with a lot of data
Very similar to a key-value database
Main difference is that you can actually see (and QUERY for) the values

B) DOCUMENT-STORE DBs: MongoDB

C) GRAPH DBs
Graph databases focus on modelling the structure of the data
Inspired by Euler’s graph theory, G=(E,V)
Motivated mainly by social media networks

D) COLUMN STORE
Column data is saved together, instead of row data
Super useful for data analytics
Inspired by Google BigTable

D) COLUMN STORE: Facebook’s Cassandra

E) OTHER DATABASES
Many other NoSQL databases exist that do not fall in these four categories:
● Text search databases (e.g. ElasticSearch)
● XML databases
● ...

Comparison of NoSQL categories

Popularity
https://db-engines.com/en/ranking

So, which one to use?
It depends
In general, RDBMS is great for ensuring data consistency, when data validity is
very important (think financial transactions and similar corporate applications)
NoSQL is great ensuring high availability and speed rather than validity
NoSQL is also great for many applications that are hard to imagine with RDBMSs:
indexing text documents, social networks, storing coordinates, etc.
Pick the right tool for the job!

Me and NoSQL: Map studio
Bachelor thesis: Use of political maps to render statistical data on countries,
provinces, areas, cities, etc.
We are drawing the maps from scratch, so we have to keep a huge number of
coordinates in our database
Typically the coordinates were stored only once, and retrieved many times. So the
READ operation is very important to be efficient, but not the WRITE one.

Me and NoSQL: Map studio
Requirements:
● Main one is response time (=how quick can the database return an
answer/the data)
● Secondary goal is horizontal scalability (=the database should allow one
easily to split the data on multiple machines)
What is less/not important:
● consistency
● concurrency

Knowledge graphs
A knowledge graph acquires and integrates information into an ontology
and applies a reasoner to derive new knowledge.
● It is a graph
● It is semantic = the meaning of the data is encoded alongside the data in the
graph, in the form of the ontology. A knowledge graph is self-descriptive, or,
simply put, it provides a single place to find the data and understand what it’s
all about.
● It is smart = the ontology allows implicit information to be inferred from explicit
data
● It is alive = it can grow and get updated

Knowledge graphs
These principles are shared in the research on Semantic Web and Linked open
data, aiming to construct a Web of things.
Each world fact is represented as a subject-predicate-object triple and stored as
Linked Open Data.
By the end of 2016 Google’s knowledge graph apparently contained 70 billion
connected facts.
The Linked Open Data cloud also contains billions of facts (our LOD Laundromat
collection at VU might be among the largest -> last counted 40B, but next run will
contain much more).

Take aways from today’s lecture
1) NoSQL fills the spectrum between flat files and relational DBs
2) NoSQL databases were introduced to fit the new trends: many users, many
queries, complex data models, diverse data structures, and cheap hardware
3) The data structure in NoSQL is often not a table, but instead a dictionary, a
tree, an array, etc.
4) Horizontal scaling, repetition of data, flexible schema
5) Classes of NoSQL DBs: graph, column, key-value, document, other
6) RDBMS or NoSQL? Pick the right tool for the job

Acknowledgements
Some slides have been copied from existing slides on Slideshare
● https://www.slideshare.net/mikecrabb/a-beginners-guide-to-nosql/
● https://www.slideshare.net/bhaskar_vk/introduction-to-nosql-databases-
47768468
● https://www.slideshare.net/RTigger/sql-vs-no-sql
https://hackernoon.com/wtf-is-a-knowledge-graph-a16603a1a25f
For an advanced presentation on NoSQL, see:
● https://www.slideshare.net/quipo/nosql-databases-why-what-and-when/34-

Thanks!
Anything you would like to talk about?

NoSQL databases

Related slideshows

More Related Content

NoSQL databases