<Insert Picture Here>

MySQL Cluster – sometimes SQL

Bernd Ocklin
 MySQL Cluster

• High Performance: Write Scalability & Low Latency

• 99.999% Availability
• Flexible Data Access Methods (SQL & NoSQL)
• Low TCO (Open Source + Commodity Hardware)
 Coming from mobile networks with
tough requirements ...
• 50, 60, 100 x 106 mobile phone users
• x 106 operations per second
• micro - millisecond latency
• no downtime, even during upgrades
• low cost
SQL/NoSQL – shared nothing
no single point of failure

Clients

native ClusterJ REST/JSON LDAP memcached SQL, JDBC,

ADO, ...

NDB API

MySQL Cluster Data Nodes

Geographic Redundancy

• synchronous replication
locally
• async replication
geographically
• master-slave or multi-
master
• automated conflict
detection and resolution
Out of the box scalability
distributed hash & data partitioning
Application
Authid (PK) Frame Iname Country
1 Albert Camus France
2 Ernest Hemingway USA
3 Johann Goethe Germany
4 Junichiro Tanizaki Japan

Authid (PK) Frame Iname Country Authid (PK) Frame Iname Country
1 Albert Camus France 2 Ernest Hemingway USA
3 Johann Goethe Germany 4 Junichiro Tanizaki Japan
Availability with sub-second failover

Application
Authid (PK) Frame Iname Country
1 Albert Camus France
2 Ernest Hemingway USA
3 Johann Goethe Germany
4 Junichiro Tanizaki Japan

Authid (PK) Frame Iname Country Authid (PK) Frame Iname Country
1 Albert Camus France 2 Ernest Hemingway USA
3 Johann Goethe Germany 4 Junichiro Tanizaki Japan
Key features

• Shared-nothing, clustered database server

Distributed • ACID-compliant relational database

Highly • Five 9s (99.999%) availability

available • Self-healing, subsecond failover

Real-time • High-load, real-time performance

performance • Predictable low latency, bounded access times

Dynamically • Incrementally scale up, out and on-line with application demands
• Linearly scale with distribution awareness
scalable

Open • Open source, multiple data access

development • High-performance APIs (C++/Java), SQL, Memcached,
REST/JSON, LDAP, Web Services
 12 years and counting … rock solid

Ericsson Network
DataBase (NDB): MySQL 5.1: MySQL Cluster 7.0:
Real-Time Disk data Scale-up (multi-
99.99% availability threaded data nodes)
Geo-Replication MySQL Cluster 7.1:
Auto-failover On-line add-node
User-defined MySQL Cluster
In-memory partitioning 4x performance
MySQL 4.1.7: Manager
Scale-out Back-NDB for LDAP
NDB/Cluster NDBINFO – real-
On-line backups integrated time monitoring
MySQL Cluster 6.1:
NDB API direct with MySQL Java Connector (ClusterJ)
255 nodes
C++ Access SQL Access
NDB/J direct Java access

1998 2004 2006 2009 2010

MySQL 5.0: MySQL Cluster 6.X:

Batched API On-line schema changes
access Enhanced NDB API
Robustness & mod_ndb REST/JSON
performance
 Who uses cluster?
Web & telecoms
• Telecoms
– Subscriber Databases (HLR/HSS)
– Service Delivery Platforms
– VoIP, IPTV & VoD
– Mobile Content Delivery
– On-Line app stores and portals
– IP Management
– Payment Gateways
• Web
– User profile management
– Session stores
– eCommerce
– On-Line Gaming
– Application Servers http://www.mysql.com/customers/cluster/
 Consolidated authentication 7.2

GRANT mrfoo

GRANT mrfoo

MySQL Cluster Data Nodes

Simplify the stack 7.2
Native memcached ndb_engine
App

App
memcached memcached

memcached
ndb_engine.so
ndb_engine.so

MySQL Cluster Data Nodes

 native cluster memcached ndb_engine

• Native cluster access through memcache protocol

• Direct write/read-through to cluster
• Using memcached 1.6 plug-in engine architecture
• Many programming languages
• Reduces system complexity
• Eliminates data inconsistencies between several
memcached instances and the db backend
 NDBAPI example

NdbTransaction * tx = ndb->startTransaction();

NdbOperation * op = tx->getNdbOperation(myTable);

op->readTuple(NdbOperation::LM_CommittedRead);

op->equal("code", code);

op->getValue("name", name);

tx->execute( NdbTransaction::Commit );
cout << "name = " << name << endl;

ndb->closeTransaction(tx);
Throughput benchmark 2011
Note: last minute benchmark before the conference to show some new numbers.

4M ops/s

3M ops/s

2M ops/s

updates
1M ops/s

4 nodes 8 nodes 16 nodes reads

Java Connector

App App App App

JPA

ClusterJPA
ClusterJPA

JDBC ClusterJ
ClusterJ

MySQL JNI

NDB API (C++)

 ClusterJ annotated interface

@PersistenceCapable(table="t_fish_food")
CREATE TABLE t_fish_food(
public interface Fish {

@PrimaryKey
int getId(); id int primary key,
void setId(int id);

@Column(name = "Fishname")
@Index(name="idx_unq_hash_name") Fishname varchar(255)
String getName(); unique index using hash,
void setName(String name);

... ...
}; ) ENGINE = ndbcluster;
 ClusterJ find() data

Employee theEmployee =
session.find(Employee.class, 988);
 ClusterJ persist()

// Create and initialise an Employee

Employee newEmployee = session.newInstance(Employee.class);
newEmployee.setId(988);
newEmployee.setFirst("John");
newEmployee.setLast("Jones");
newEmployee.setStarted("1 February 2009");
newEmployee.setDepartment(666);

// Write the Employee to the database

session.persist(newEmployee);
 mod_ndb

• Community feature
• Apache module allowing native access to cluster
• REST API speaking JSON or XML

> GET /flights/airports?iata=SFO

{iata : SFO,
Long : “San Francisco Intl Airport“,
Loc : US}
 mod_ndb and AJAX

$.get('http://localhost/demo/iata',
{ "code": value },
function(data) {
var n = JSON.parse(data);
output(n.name);
});
 SQL, memcached, ClusterJ, C++,
REST/JSON … which to choose?
• Answer: Use either or both, whatever best fits your application
– Different APIs into MySQL Cluster
– Alternative data stores
• Factors to consider:
– Performance & Scalability
– Developer skills & Familiarity with APIs
– Levels of support
– Access patterns (joins needed? Key/value sufficient?)
– Schema changes (online or schema-less)
• Mix & Match!
– MySQL Cluster allows the same data to be accessed simultaneously
through SQL & NoSQL interfaces
 SQL, memcached, ClusterJ, C++,
REST/JSON … which to choose?
SQL
•
Using a standard mod_ndb
•
Joins & complex queries •
REST/JSON
•
Relational model •
HTML
Memcached •
using apache
•
simple to use API
•
key/value
•
driver for many languages
•
ideal as e.g. PHP proxy
ClusterJ
•
simple to use Java API
•
Web & telco C++
•
Object Relational Mapping •
knowledged developer
•
native & fast access to cluster •
super low latency / real-time
Sometimes SQL!

• Allows most complex

business questions
• Standard – most know it
• But cluster was
traditionally not optimal
with joins
SQL, JDBC, SQL, JDBC,
ADO, ... ADO, ...

MySQL Cluster Data Nodes

 Scaling distributed joins in cluster! 7.2

Traditionally
• Very fast primary key operations
• Certain optimized clauses (e.g. IN(....))
• Distributed joins were hard to scale

New in 7.2
• Pushing the execution down into the storage layer,
greatly reduces network trips
• Makes joins scale and up to 40x faster
• expand the use of MySQL Cluster into a broader
range of services and applications
COMPANY OVERVIEW USER PERSPECTIVE
• Division of Docudesk “MySQL Cluster exceeds our requirements for low
• Deliver Document Management SaaS latency, high throughput performance with
continuous availability, in a single solution that
minimizes complexity and overall cost.”
CHALLENGES / OPPORTUNITIES -- Casey Brown, Manager of Dev & DBA Services,
• Provide a single repository for customers to Docudesk
manage, archive, and distribute documents
• Implement scalable, fault tolerant, real time
data management back-end RESULTS
• PHP session state cached for in-service • Successfully deployed document
personalization management solution, eliminating paper
• Store document meta-data, text (as trails from legal processes
BLOBs), ACL, job queues and billing data • Integrate caching and database into one
• Data volumes growing at 2% per day layer, reducing complexity & cost
• Support workload with 50:50 read/write
ratio
SOLUTION • Low latency for real-time user experience
• MySQL Cluster deployed on EC2 and document time-stamping
• Continuous database availability
Star Schema Q1.1 with distributed joins

• Likely never as fast

as specilized
databases
• But only moderatly
slower with parallel
50k updates / sec
make a unique
combination.
 Download 7.2 Dev Milestone now

Download the software:

7.1 GA: www.mysql.com/downloads/cluster/
7.2 DMR: dev.mysql.com/downloads/cluster/
Memcached: labs.mysql.com
Blogs on latest developments: clusterdb.com, planet.mysql.com
MySQL Cluster Quick Start Guides
www.mysql.com/products/database/cluster/get-started.html#quickstart
MySQL Cluster 7.1, Architecture and New Features
www.mysql.com/why-mysql/white-papers/mysql_wp_cluster7_architecture.php
MySQL Cluster on the Web
www.mysql.com/products/database/cluster
Thank you!
The presentation is intended to outline our general
product direction. It is intended for information
purposes only, and may not be incorporated into any
contract. It is not a commitment to deliver any
material, code, or functionality, and should not be
relied upon in making purchasing decisions.
The development, release, and timing of any
features or functionality described for Oracle’s
products remains at the sole discretion of Oracle.