Who am I?

• Founder of Astrails

• 19 years of industry experience

• Building Web Apps since 2005

Vitaly Kushner

Wednesday, June 16, 2010

NoSQLwhen, why and how?

Vitaly Kushnerastrails.com

Wednesday, June 16, 2010

Google

Wednesday, June 16, 2010

Google

Yahoo

Wednesday, June 16, 2010

Google

Yahoo

Amazon

Wednesday, June 16, 2010

Google

Yahoo

Amazon

Facebook

Wednesday, June 16, 2010

Google

Yahoo

Amazon

Facebook

Twitter

Wednesday, June 16, 2010

Google

Yahoo

Amazon

Facebook

Twitter

Digg

Wednesday, June 16, 2010

Google

Yahoo

Amazon

Facebook

Twitter

Rackspace

Digg

Wednesday, June 16, 2010

Google

Yahoo

Amazon

Facebook

Twitter

Rackspace

Digg

LinkedIn

Wednesday, June 16, 2010

Google

Yahoo

Amazon

Facebook

Twitter

Rackspace

Digg

LinkedInEverybody

Wednesday, June 16, 2010

NoSQL

NoSQL

NoSQLNoSQL

NoSQL

NoSQL

NoSQL

NoSQL

NoSQL

NoSQL

Wednesday, June 16, 2010

WTF is NoSQL?and why should you care?

Wednesday, June 16, 2010

Non relational

Wednesday, June 16, 2010

Non relationalDocument based

Wednesday, June 16, 2010

Non relationalDocument based

Key-Value store

Wednesday, June 16, 2010

Non relationalDocument based

Key-Value store

column-based

Wednesday, June 16, 2010

Non relationalDocument based

Key-Value store

Graph DB

column-based

Wednesday, June 16, 2010

Non relationalDocument based

Key-Value store

Distributed

Graph DB

column-based

Wednesday, June 16, 2010

Non relationalDocument based

Key-Value store

Schema-less

Distributed

Graph DB

column-based

Wednesday, June 16, 2010

Non relationalDocument based

Key-Value store

Schema-less

Distributed

BASE is not ACID Graph DB

column-based

Wednesday, June 16, 2010

Why & When

Wednesday, June 16, 2010

Massive Data Volume100K servers in a cluster

Wednesday, June 16, 2010

Massive Data Volume100K servers in a cluster

Twitter: 7+T/day

Wednesday, June 16, 2010

High query workloadMongoDB: 8M operations/sec

Wednesday, June 16, 2010

Flexible Schemaon the fly schema changes

Wednesday, June 16, 2010

Massive Scale

Wednesday, June 16, 2010

Availability

Wednesday, June 16, 2010

Everyone wantAvailability

Wednesday, June 16, 2010

RDBMScan deliver

Wednesday, June 16, 2010

high price

Wednesday, June 16, 2010

Not ACID anymore

Wednesday, June 16, 2010

• Consistent

• Available

• Partition tolerant

CAP theoremPick two

Wednesday, June 16, 2010

ScaleHow?

Wednesday, June 16, 2010

Throw hardware money at it!

Wednesday, June 16, 2010

Par-ti-tion

Wednesday, June 16, 2010

MySQL + Memcached=

“square wheel” cassandra

Wednesday, June 16, 2010

How?

Wednesday, June 16, 2010

Which one?

• document based

• column or key-value store

• advanced storage schemas

Wednesday, June 16, 2010

Cassandra

• built by Facebook

• very high write throughput

• OLTP

• automatic horizontal scaling

• no single point of failure

Wednesday, June 16, 2010

HBase

• Apache project

• Consistent

• Optimized for analytics (OLAP)

• Has single point of failure

Wednesday, June 16, 2010

MongoDB

• probably easiest to move to from SQL

• document based

• on-demand queries

• automatic sharding

• no single-node durability

Wednesday, June 16, 2010

CoachDB

• document based

• map-reduce javascript querying/filtering.

• has some replication and scaling problems

Wednesday, June 16, 2010

REDIS

• key-value store

• advanced data types: list, set

• atomic operations

Wednesday, June 16, 2010

Schema

Wednesday, June 16, 2010

Users: {

vitaly: {

email: vitaly@astrails.com,

company: astrails,

password: secret

},

michael: {

email: michael@astrails.com,

company: astrails,

password: superduper

},

...

}

Wednesday, June 16, 2010

UsersByEmail: { "vitaly@astrails.com": "vitaly", "michael@astrails.com": "michael", ...}

Wednesday, June 16, 2010

Migrations

Wednesday, June 16, 2010

Start Slow

Wednesday, June 16, 2010

NoSQL can helpYou

Wednesday, June 16, 2010

WTF is NoSQL?

Vitaly Kushnerastrails.com

@astrails @vkushner

Q & A

Wednesday, June 16, 2010