Relational or Not Full

8/14/2019 Relational or Not Full

1/98

Relational

vs.Non-RelationalJosh BerkusPostgreSQL Experts Inc.

SCALE 2010


2/98

2003:

MySQL

PostgreSQL

FireBird BerkeleyDB

Derby

HSQLDB SQLite


3/98

2003:

MySQL

PostgreSQL

FireBird BerkeleyDB

Derby

HSQLDB SQLite


4/98

2010

MySQL

PostgreSQL

FireBird BerkeleyDB

Derby

HSQLDB SQLite


5/98

NoSQLmovement


6/98

All

non-relational

databases


7/98

Are notthe same


8/98

Graph

Key-value

Document

Hierarchical

Distributed

Neo4JHyperGraphDBJena

CouchDBBerkeleyDB-XMLS olr

MemcachedTokyo Cabinet

db4oRIAK

CassandraHypertableMyS QL NDB

MongoDB


9/98

All

relationaldatabases


10/98

Are notthe same


11/98

Embedded

M PP

OLTP

Streaming

C -S tore

S QLiteFirebirdHSQL

PostgreS QLMySQLOracleS QL S erver

TeraDataGreenplum

Aster

LucidDBMonetDB

S treambase

Truviso


12/98

NoFinsNoFins


13/98


14/98

NoSQLmovement


15/98

Mythbust #2


16/98

revolutionary


17/98

There


18/98

are


19/98

no


20/98

new


21/98

database


22/98

designs


23/98

There are only newimplementations

andcombinations


24/98

A database storing application-friendly formatted objects, each

containing collections ofattributes which can be searched

through a document ID, or thecreation of ad-hoc indexes asneeded by the application.


25/98

CouchDB, 2007

Pick, 1965


26/98

CouchDB, 2007

embeddable PickJSON storage

REST APImap/reduce


27/98

revolutionary


28/98

revolutionary


29/98

renaissanceof

non-relationaldatabases


30/98

Mythbust #3


31/98

non-relational

databasesare toys


32/98

Bigtable

Google


33/98

Dynamo

Amazon


34/98

Memcached

FaceBook


35/98

Pick, Cach

US Vetrans'

Administration


36/98

Mythbust #4


37/98

Relational

databaseswill become

obsolete


38/98

Three decades past, the relational

empire conquered the hierarchicalhegemony. Today, an upstartchallenges the relational empire's

dominance ...

--Philip Wadler, KeynoteVLDB, Rome, September 2001

XML Databases

2001


39/98

Anyone remember

XML databases?


40/98

No?


41/98

What happened?


42/98

established relationaland non-relational

databaseshybridized XML


43/98


44/98

Mythbust #5


45/98

Relational databases

are for when you needACID transactions.


46/98

Transactions

Relational


47/98

Robust Transactions without

Relationality:

SQL Without Transactions:

BerkeleyDB

Amazon Dynamo

MySQLMyISAMMS Access


48/98

Mythbust #6


49/98


50/98

You


51/98

do


52/98

not


53/98

have


54/98

to choose


55/98

onedatabase.


56/98

Choosethe database systemwhich fi ts your

currentapplication goals.

or ...

Use more than one


57/98

together

MySQL + Memcached

PostgreSQL +CouchDB

or ...

Use a Hybrid


58/98

Use a Hybrid

MySQL NDB

PostgreSQL Hstore

HadoopDB


59/98

But what aboutrelational

vsnon-relational?

Relational OLTP


60/98

Databases* Transactions: more mature support including multi-statement

Constraints: enforce data rules absolutely Consistency: enforce structure 100%

Complex reporting: keep management happy!

Vertical scaling (but not horizontal)* mature ones, anyway

SQL vs. Not SQL


61/98

SQL vs. Not SQL

SQL promotes: portability

managed changes over time multi-application access many mature tools

SQL vs. Not SQL


62/98

SQL vs. Not SQL

But SQL is a full programming language,and you have to learn it to use it.

SQL vs. Not SQL


63/98

SQ s ot SQ

No-SQL allows: programmers as DBAs

no impedance fast interfaces fast development


64/98


65/98

Immortal Data

your data hasa lifeindependent

of this specifi c

application implementation


66/98

How do I

choose?


67/98

Defi ne the problem

you are tryingto solve

I need a database for my blog

I need to add 1000's of objects per second on a low-d d i


68/98

end device.

I need my database to enforce business rules acrossseveral applications.

I want my application to be location-sensitive.

I need to cache data an access it 100K times per

second. I need to produce summary reports across 2TB of

data.

I have a few hundred government documents I need

to serve on the web and mine I need to know who-knows-who-knows-who.

I need to data mine 1000's of 30K bug reports perminute.

Defi ne the featurest ll d


69/98

you actuallyneed many connections

multi-server scalability

complex query logic APIs

redundancy

data integrity schema/schemaless

data mining


70/98

fi t the database

to the task


71/98

I need a database

for my blog

Use anything!


72/98

y g

MySQL PostgreSQL

MongoDB

SQLite CouchDB

Flatfi les

DBaseIII Something you wrote yourself


73/98

I need my databaseto unify several

applications andkeep them consistent.

PostgreSQL


74/98

OLTPSQL-Relational Database

It's not just a database: it's a developmentplatform


75/98

Postgres 9Postgres 9

alpha out now!alpha out now!


76/98

I need my application

to be location-aware.

PostGIS


77/98

Geographic RelationalDatabase


78/98


79/98

I need to store1000's of event objects

per second onembedded hardware.

db4object


80/98

Embedded Key-ValueStore

db4object


81/98

German Train System Insert 1000's of objects per second

Low-end embedded console computer

Simple access in native programming language(Java, .NET)

compromise: embedded SQL database:SQLite


82/98

I need to access100K objects per

secondover 1000's of

connections.

memcached


83/98

Distributed In-MemoryKey-Value Store

memcached


84/98

Use: public website Used for caching 1000's of serialized objects

per second

Available for 100000's of requests per second

across 1000's of connections Cache each object only once per site

Supplementsa relational database

Alternatives: Redis, TokyoTyrant


85/98

I need to producecomplex summary

reportsover 2TB of data.

LucidDB


86/98

RelationalColumn-Store

LucidDB


87/98

For reporting and analysis Large quantities of data (TB)

Complex OLAP & analytics queries

Business intelligence complimentsa transactional database


88/98

I have 100's ofgovernment documentsI need to

serve on the weband mine for data.

CouchDB


89/98

Document Store

CouchDB


90/98

1.CividDB Project2.Storing lots and lots of government documents

3.Don't know what's in them

4.Don't know how they are structured

5.Store them, fi gure out structure later by datamining.

It's also good for web sites!

I have a social


91/98

application andI need to know

who-knows-who-knows-

who-knows-who-knows-who.

Neo4j


92/98

Graph Database

Neo4j


93/98

Social Network Website

6 degrees of separation

you may also like

type and degrees of relationship

I get 1000's of


94/98

I get 1000 s of30K bug reports

per minute

and I needto mine them for

trends.

Hadoop


95/98

Massively Parallel Data Mine

HadoopM i b d f d


96/98

Massive bug report data feed

1000's of bug reports per minute

Each bug report 2-45K of data

Need to extract trends and correlate inexactdata

Summarize in daily & weekly reports

ConclusionDiff t d t b t d b tt t


97/98

Different database systems do better atdifferent tasks.

every database feature is a tradeoff

no database can do all things well

Relational vs. non-relational doesn't matter pick the database(s) for the project or the task

Questions?P t SQL P j t P t SQL B th


98/98

PostgreSQL Project

www.postgresql.org

[email protected]

PostgreSQL Experts

www.pgexperts.com

www.pgexperts.com/documents.html

Open Source Database Survey

Selena Deckleman

Open Source Database Survey:www.ossdbsurvey.org

PostgreSQL Boothtrade show fl oor

PostgreSQL 9.0BOF, Tonight 7pm

Copyright 2010 Josh Berkus, distributable under the creative commons attribution license,except for 3rd-party images which are property of their respective owners.Special thanks to Selena Deckelman for the bunnies image and forthe Open Source Database survey.

Documents

Relational or Not Full