Max Neunhöffer presents on the future of NoSQL databases and argues multi-model databases will become the standard. He discusses different NoSQL data models like document stores, key-value stores, graph databases and column-oriented databases. He advocates the benefits of a polyglot persistence approach but notes the disadvantages of managing multiple databases. Max introduces ArangoDB as a multi-model database that supports documents, graphs and key-value in a single database to provide the benefits of polyglot persistence without the disadvantages. He provides examples of how ArangoDB has been used and outlines its features including queries, extensibility and horizontal scalability. Max predicts that in five years, the default approach will be to use a

1. Is multi-model the future of
NoSQL?
Max Neunhöffer
SouthBay.NET Meetup, 5 March 2015
www.arangodb.com

2. Max Neunhöffer
I am a mathematician
“Earlier life”: Research in Computer Algebra
(Computational Group Theory)
Always juggled with big data
Now: working in database development, NoSQL, ArangoDB
I like:
research,
hacking,
teaching,
tickling the highest performance out of computer systems.
1

3. ArangoDB GmbH
triAGENS GmbH offers consulting services since 2004:
software architecture
project management
software development
business analysis
a lot of experience with specialised database systems
have done NoSQL, before the term was coined at all
2011/2012, an idea emerged:
to build the database one had wished to have all those years!
development of ArangoDB as open source software since 2012
ArangoDB GmbH: spin-off to take care of ArangoDB (2014)
2

4. Document and Key/Value Stores
Document store
A document store stores a set of documents, which usually
means JSON data, these sets are called collections. The
database has access to the contents of the documents.
each document in the collection has a unique key
secondary indexes possible, leading to more powerful queries
different documents in the same collection: structure can vary
no schema is required for a collection
database normalisation can be relaxed
Key/value store
Opaque values, only key lookup without secondary indexes:
=⇒ high performance and perfect scalability
3

5. Graph databases
Graph database
A graph database stores a labelled graph. Vertices and
edges can be documents. Graphs are good to model
relations.
graphs often describe data very naturally (e.g. the facebook
friendship graph)
graphs can be stored using tables, however, graph queries
notoriously lead to expensive joins
there are interesting and useful graph algorithms like “shortest
path” or “neighbourhood”
need a good query language to reap the benefits
horizontal scalability is troublesome
graph databases vary widely in scope and usage, no standard
4

6. Column-oriented data stores
Column-oriented data astores
A column-oriented database stores tables but “keeps
columns together” rather than rows.
access to a whole column is fast
sparse rows are handled efficiently
particularly good for certain types of data analysis
often implemented in a key/value-like fashion
row access can be slow
columns have homogeneous data, so compression works well
prominent examples: C-Store and Cassandra
5

7. Massively parallel: map-reduce and friends
The area of massively parallel
A massively parallel database can use thousands of servers
distributed all over the world and still appears as a single
service.
Humongous data capacity and very high read/write
performance
examples are Apache Cassandra, Apache Hadoop, Google’s
Spanner, Riak and others
these systems have important use cases, in particular in the
analytic domain
query capabilities are somewhat limited like for example only
“map/reduce”
⇒ good horizontal scalability at the cost of reduced query flexibility
6

8. Polyglot Persistence
Idea
Use the right data model for each part of a system.
For an application, persist
an object or structured data as a JSON document,
a hash table in a key/value store,
relations between objects in a graph database,
a homogeneous array in a relational DBMS.
If the table has many empty cells or inhomogeneous rows, use
a column-oriented database.
Take scalability needs into account!
7

9. A typical Use Case — an Online Shop
We need to hold
customer data: usually homogeneous, but still variations
=⇒ use a relational DB: MySQL
product data: even for a specialised business quite
inhomogeneous
=⇒ use a document store:
shopping carts: need very fast lookup by session key
=⇒ use a key/value store:
order and sales data: relate customers and products
=⇒ use a document store:
recommendation engine data: links between different entities
=⇒ use a graph database:
8

10. Polyglot Persistence is nice, but . . .
Consequence: One needs multiple database systems in the persis-
tence layer of a single project!
Polyglot persistence introduces some friction through
data synchronisation,
data conversion,
increased installation and administration effort,
more training needs.
Wouldn’t it be nice, . . .
. . . to enjoy the benefits without the disadvantages?
9

11. The Multi-Model Approach
Multi-model database
A multi-model database combines a document store with a
graph database and is at the same time a key/value store.
Vertices are documents in a vertex collection,
edges are documents in an edge collection.
a single, common query language for all three data models
is able to compete with specialised products on their turf
allows for polyglot persistence using a single database
queries can mix the different data models
can replace a RDMBS in many cases
10

12. Why is this possible at all?
Document stores and key/value stores
Document stores: have primary key, are key/value stores.
Without using secondary indexes, performance is nearly as
good as with opaque data instead of JSON.
Good horizontal scalability can be achieved for key lookups.
11

13. Why is this possible at all?
Document stores and graph databases
graph database: would like to associate arbitrary data with
vertices and edges, so JSON documents are a good choice.
A good edge index, giving fast access to neighbours.
This can be a secondary index.
Graph support in the query language.
Implementations of graph algorithms in the DB engine.
12

14. A Map of the NoSQL Landscape
Transaction Processing DBs
Analytic processing DBs
Map/reduce
Column Stores
Extensibility
Documents
Massively distributed
Graphs
Structured
Data
Key/Value
Complex queries
13

15. Use case: Aircraft fleet management
One of our customers uses ArangoDB to
store each part, component, unit or aircraft as a document
model containment as a graph
thus can easily find all parts of some component
keep track of maintenance intervals
perform queries orthogonal to the graph structure
thereby getting good efficiency for all needed queries
14

16. Use case: Family tree management
For genealogy, the natural object is a family tree.
data naturally comes as a (directed) graph
many queries are traversals or shortest path
but not all, for example:
“all people with name James” in a family tree, sorted by birthday
“all family members who studied at Berkeley”, sorted by
number of children
quite often, queries mixing the different models are useful
15

17. Use case: knowledge bases
encode nearly arbitrary knowledge
often produced by machine learning
queried in very complex ways by expert systems
often in connection to an inference engine
need linked data with lots of associations
typical queries have unpredictable path length, thus graph
queries shine
nevertheless, often queries orthogonal to the links are needed
16

18. Recently: Key/Value stores adding other models
(by Basho), originally a key/value store, adds support for
documents with their 2.0 version (late 2014)
(sponsored by Pivotal), originally an in-memory
key/value store, has over time added more data types and
more complex operations
FoundationDB (by FoundationDB) is a key/value store, but is
now marketed as a multi-model database by adding additional
layers on top
OrientDB (by Orient Technologies) started as an object
database and nowadays calls itself a multi-model database
17

19. Recently: DataStax acquired Aurelius
In February 2015, DataStax (commercialised version of Cassan-
dra (column-oriented)), announced the acquisition of Aurelius, the
company behind TitanDB (a distributed graph database on top of
Cassandra).
In their own words:
“Bringing Graph Database Technology To Cassandra.”
“Will deliver massively scalable, always-on graph database
technology.”
“Will simplify the adoption of leading NoSQL technologies to
support multi-model use case environments.”
18

20. Recently: MongoDB 3.0 adds pluggable DB engine
is one of the most popular document stores.
In February 2015, they announced their 3.0 version, to be released
in March, featuring
a pluggable storage engine layer
transparent on-disk compression
etc.
This indicates their interest to support more data models than “just
documents”.
It will be very interesting indeed to see if and how they extend their
query-language . . .
19

21. is a multi-model database (document store & graph database),
is open source and free (Apache 2 license),
offers convenient queries (via HTTP/REST and AQL),
memory efficient by shape detection,
uses JavaScript throughout (Google’s V8 built into server),
API extensible by JavaScript code in the Foxx framework,
offers many drivers for a wide range of languages,
is easy to use with web front end and good documentation,
enjoys good professional as well as community support
and has sharding since Version 2.0.
20

22. Configurable consistency
ArangoDB offers
atomic and isolated CRUD operations for single documents,
transactions spanning multiple documents and multiple
collections,
snapshot semantics for complex queries,
very secure durable storage using append only and storing
multiple revisions,
all this for documents as well as for graphs.
In the near future, ArangoDB will
implement complete MVCC semantics to allow for lock-free
concurrent transactions
and offer the same ACID semantics even with sharding.
21

23. Replication and Sharding — horizontal scalability
Right now, ArangoDB provides
easy setup of (asynchronous) replication,
which allows read access parallelisation (master/slaves setup),
sharding with automatic data distribution to multiple servers.
Very soon, ArangoDB will feature
fault tolerance by automatic failover and synchronous
replication in cluster mode,
zero administration by a self-reparing and self-balancing
cluster architecture,
full integration with Apache Mesos and Mesosphere.
22

24. Powerful query language: AQL
The built in Arango Query Language AQL allows
complex, powerful and convenient queries,
with transaction semantics,
allowing to do joins,
with user definable functions (in JavaScript).
AQL is independent of the driver used and
offers protection against injections by design.
For Version 2.3, we have reengineered the AQL query engine:
use a C++ implementation for high performance,
optimise distributed queries in the cluster.
23

25. Extensible through JavaScript and Foxx
The HTTP API of ArangoDB
can be extended by user-defined JavaScript code,
that is executed in the DB server for high performance.
This is formalised by the Foxx microservice framework,
which allows to implement complex, user-defined APIs with
direct access to the DB engine.
Very flexible and secure authentication schemes can be
implemented conveniently by the user in JavaScript.
Because JavaScript runs everywhere (in the DB server as well
as in the browser), one can use the same libraries in the
back-end and in the front-end.
=⇒ implement your own micro services
24

26. The Future of NoSQL: My Observations
I observe
2 decades ago the most versatile solutions eventually
dominated the relational DB market
(Oracle, MySQL, PostgreSQL),
the rise of the polyglot persistence idea
a trend towards multi-model databases
specialised products broadening their scope
even relational systems add support for JSON documents
devOps gaining influence (Docker phenomenon)
25

27. The Future of NoSQL: My Predictions
In 5 years time . . .
the default approach is to use a multi-model database,
the big vendors will all add other data models,
the NoSQL solutions will conquer a sizable portion
of what is now dominated by the relational model,
specialized products will only survive, if they find a niche.
26

28. Links
https://www.arangodb.com
http://guesser.9hoeffer.de:8000
https://github.com/ArangoDB/guesser
https://github.com/triAGENS/ArangoDB-NET
27