Scaling out Web Applications with Microsoft SQL Azure DatabaseDavid RobinsonSenior Program ManagerMicrosoft Corporation

SVC06

Session Overview> Brief SQL Azure Recap> Why scale your workload> Application Scaling techniques> Considerations when scaling out> Integrated Scale-Out Support

SQL Azure Database

Customer Value PropsSelf-provisioning and capacity on demandSymmetry w/ on-premises database

platformAutomatic high-availability and fault-

toleranceAutomated DB maintenance

(infrastructure)Simple, flexible pricing – “pay as you

grow”

Browser

MSDatacenter

SOAP/RESTHTTP/S

SQL Azure Database (Windows Azure Compute)

Windows Azure Compute

T-SQL (TDS)

App Code(ASP.NET)

SQL Server Reporting Server

(on-premises)

Astoria/REST - EDMHTTP/S

- AD Federation (LiveId /.Net Svcs ACS)

SQL AzureDatabase

T-SQL (TDS)

Relational database service

• SQL Server technology foundation• Highly symmetrical• Highly scaled

Database “as a Service” – beyond hosting

Highly scaled out relational database as a service

Why scale your application> Increase an application’s ability to

process and store data> Usually because of heavy resource

consumption> Increased workloads> Increased CPU / IO requirements> Increased storage requirements

Scale-out SupportDatabase and workload partitioning is a classic technique for harnessing the power of many computers to achieve:

• Better price/performance • Levels of throughput not possible with a single machine

Classic scale-out challenges addressed by SQL Azure: Highly available service on top of commodity hardware Zero administration cost of cluster HW, OS and SW Access to elastic pool of resources Pay as you grow No-friction provisioning of databases Create databases without ever running out of HW

Real World ScenariosKelly Blue Book

>Provider of vehicle valuation data>> 13 million visitors to their site per month>2.5 GB Database Size>Data refreshed every week>Replicate data across 5 databases for

increased perf>Increase/Decrease database count based on

demand

customerChris AuldDirector of Strategy & InnovationIntergen

SQL Azure

Castellan

Castellan Venue

DBCastellan

Venue DB’s

Venue 1Partition(

s)

Castellan Venue

DBCastellan

Venue DB’s

Venue 2 Partition(

s)

Castellan Venue

DBCastellan

Venue DB’s

Venue N Partition(

s)One application DB, many venue DB’s – each partitioned in to many parts (40+)

...

Azure Roles

http:// TicketDirect .* Dynamic Worker(tasks

uploaded as blobs)

Partitioner Worker

Azure StorageQueues for communication between clients and roles

-- - --- - --- - --- - -

Tables to record server & partition information

Blobs to store web and worker role resources

Client Applications

Castellan.old (VB6)

Castellan.Azure• Box Office sales• Ticket Printing • System Administration • Venue/Event Management• Partitioning

.Net Service Bus

WCF

On PremiseSQL Server

Castellan Venue

Distributed Cache Worker

MemCache

TicketDirect Architecture

What are your application’s requirements?Storage and Transactional throughput

Storage RequirementsLow High

Tran

sact

iona

l Req

uire

men

tsLo

wH

igh

• Single Database• No Partitioning

• Partitioned Data• Partitioning Based on

Application Requirements (Storage)

• Partitioned Data• Partitioning based on

Application Requirements (IOPS)

• Partitioned Data• Partitioning based on

Application Requirements (IOPS, Storage or both)

Scale Out Patterns> Multiple ‘standard’ scale out patterns> Range – break range into chunks

> Ranges can be variable in size> Good for range based queries> Can suffer from hotspots depending on

workload> Hashing – apply hash to partitioning keys

> Good for distributing values > Poor for range queries (needs full fan-out)> Need to accurately estimate workload

requirements

Fabrikam FishOur Scenario> Online reseller of exotic aquarium and

pond fish> Peak sales periods

> Nov – Dec> May – July

> 726,000 different varieties of fish

Fabrikam’s Schema

Product> Stock table can become

hot> Range Partitioned> Utilizes 1gb databases

Product Implementation

code walkthrough

Fabrikam’s Schema

Product> Stock table can become

hot> Range Partitioned> Utilizes 1gb databases> Certain reference data

replicated to Customer Order databases

Customer Order> Need to evenly distribute

load > Hash Partitioned> Start with 1gb databases> Move to 10gb databases

should additional capacity required

Customer Order Implementation

code walkthrough

Application PartitioningConsiderations> Schema Design & Management> Reference Data & Synchronization> Request Routing> Fan Out Queries

Schema Design & Management> Schema design should avoid cross

database joins and transactions> DDL should be upgrade resilient and

idempotent> Application code should either:

> Support multiple schemas during an upgrade

> Stop processing requests during an upgrade

Resilient DDLIF OBJECT_ID('dbo.tbl_directory') IS NULLBEGIN

CREATE TABLE dbo.tbl_directory(

…)

ENDGOIF NOT EXISTS

(SELECT * FROM SYSCOLUMNSWHERE ID = OBJECT_ID('dbo.tbl_directory') AND NAME = 'last_updated')

ALTER TABLE dbo.tbl_directory ADDlast_updated DATETIME NULL

GO

Request Routing> Application needs to be “Partition”

aware> Partition is a unit of transactional

consistency> Multiple partitions are independent of

each other> Function is used to locate a partition

Reference Data & Synchronization> Avoid Cross Database Joins> Replicate Reference Data for

Performance Reasons> Option 1 Sync Framework

> Works Great!!!> Host in Azure Worker Role

> Option 2 – Manual Update Scripts> Manual operations task> How frequent do these updates occur?> If frequent is it really reference data?

Fan Out Queries> Issue query to multiple databases in

parallel and aggregate the results> Use multiple connections and

multithreading for increased performance

> Useful in many scenarios

>>FUTUREComing Scale-out Support

Better tooling for developers and administrators:o Dynamic database partition splitso Ability to merge database partitionso Improved schema management across database

partitions

Database Features for scale-out application patterns:o Connection managemento Fan-out query support

Summary> Brief SQL Azure Recap> Scaling out provides virtually

unlimited storage and better performance

> Range and Hash Scale-Out Patterns> Considerations when scaling out

> Schema Design and Management> Request Routing> Reference Data> Fan Out Queries

> Future Scale-Out Support

Must See SQL Azure Sessions> Microsoft SQL Azure Database: Under

the Hood (SVC12)> The Future of Database Development

with SQL Azure (SVC27)> Using the Microsoft Sync Framework

to Connect Apps to the Cloud (SVC23)

YOUR FEEDBACK IS IMPORTANT TO US! Please fill out session evaluation

forms online atMicrosoftPDC.com

Learn More On Channel 9> Expand your PDC experience through

Channel 9

> Explore videos, hands-on labs, sample code and demos through the new Channel 9 training courses

channel9.msdn.com/learnBuilt by Developers for Developers….

© 2009 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.