This document provides an overview of the Windows Azure platform. It discusses how Windows Azure is designed from the ground up for massive scale across global data centers. It also describes how Windows Azure allows applications to scale individual parts up and down as needed. Finally, it outlines the core building blocks of Windows Azure including compute, storage (blobs, tables, queues), and networking.

1. Understanding the
Windows Azure Platform
David Gristwood
Application Architect, Microsoft
blogs.msdn.com/david_gristwood

2. Why a Cloud Platform?
 Reduce capital & operations costs
 Simplify application deployment & management
 Simplify scaling to internet scale
 Cost effectively handle peak loads
 Focus on new functionality & not infrastructure

3. The Azure Philosophy

4. Windows Azure is about Scale
• Windows Azure is
designed from the ground
up for true Internet scale
• Global presence
• Ride the “network effect”
• LinkedIn: 16 months for first
million users, 11 days for most
recent million
• Facebook: 5 years to hit 250m
users, 8 months to double
• Scale up and scale down
• Supports the “Scale Fast,
Fail Fast” model
www.flickr.com/photos/ScroffTheBad

5. Application Characteristics
“Scale & Elasticity”
• Need massive scale • target.com • walmart.com
• Social networking, web 2.0 • toysrus.com • barnesandnoble.com
• Need high reliability
• LOB, software as a service
• Have variable load / unpredictable ~4x normal load
lifetime (Holiday shopping)
• Ticketing, marketing
• Parallel processing
• Financial applications
• Don’t fit in well with a company’s
data center
• Avoid IT dept, joint ventures Jan 2009 Jan 2010
• Benefit from external storage
• Archive Source: Alexa
http://www.flickr.com/photos/39919537@N06/3666813048/

6. Helps your Systems Scale
• Most architecture is
hidden inside code
• Windows Azure
encourages you to
define your application
into constituent parts
• Windows Azure can
scale the parts of your
application as
appropriate
www.flickr.com/photos/39818355@N00/2128626176

7. Flexible Architecture
• Systems evolve over
time
• Cloud is not an “all
or nothing”
proposition
• Windows Azure
makes it easier to
extend to the cloud
www.flickr.com/photos/tomkpunkt/3695066577

8. Opening up the Cloud
• Allow developers to
apply their existing
skills to the cloud
• Support for .NET,
Ruby, PHP, Java
• Interoperability with
any platform, tools or
technology
www.flickr.com/photos/h19/2306213399/in/set-72157594224557858

9. Natural Platform Evolution
• Windows Server 2008
• Virtualisation
• Windows Services,
such as IIS, logging,
diagnostics, etc
• Visual Studio & .NET

10. Flexible Business Model
• No up front
hardware
purchase
• Pay as you go
costing model
• Ideal in current
economical
climate
www.flickr.com/photos/cleebster/2508660124

11. Azure Adoption

12. Azure is Live…..
http://www.liveonazure.com/

13. Managing Data Centric Applications with Capacity Bursts
Challenges Today
“Capacity Bursting“
Elastic scale – database as a service
Pay as you grow and shrink
Easy to provision and manage
Capacity
Average Time
Usage database
Compute
# of Hr’s
Average Usage No hardware, no manual database
administration required
Promotions, events, ticket selling
businesses are “bursts – bound” by
nature
Capacity constraints limit business agility
High costs of entry into new business
Difficult to roll out extra capacity quickly
Idle capacity “off-bursts” is cost
prohibitive

14. Data Centers

15. Northern
North Central Europe
USA
Eastern Asia
Western
Europe
South Central
USA
Southeast
Asia

16. Dublin
Chicago
Amsterdam
San
Antonio
Hong Kong
Singapore
Hosting locations within 100ms of
100ms of the customer
Regional hosting
At least one hosting location can
can be reached within 100ms, but
but not two
200ms Latency from 2
No points to test from or greater
regional hosting greater than 200ms latency

17. Microsoft has run Online Services since 1994
1st MSFT datacentre built in 1989
Running Online Services 24x7x365
Global service delivery to 59 markets/36 languages
∙ 30B Live ID authentications/month ∙ 2B Bing queries/month
∙ 10B MSN page views/month ∙ 240B Messenger messages/month
Global Infrastructure to Run Your Services World Wide
Quincy, WA: Approx 500K sq ft, hydro-powered
San Antonio, TX: Approx 475K sq ft, recyclable water
Chicago, IL: Approx 700K sq ft, water economization
Dublin, Ireland: Approx 303+K sq ft, air economization
http://www.globalfoundationservices.com/

18.  One of world’s largest datacentres using
containers, on-line on July 2009
 Cost $500 million,
 700k sq ft facility (approx 16 football
fields)
 30 MegaWatts today, 60 MW for future
use
 Security
 ISO 27001:2005 Accreditation - Security
Management
 SAS 70 Type I and II Attestations - Audit
standard for our internal controls system
 Containers
 40 ft containers with 1800-2500 servers
 10 x density traditional datacentres
 Optimizes efficiency, reduces wastes &
carbon footprint
 Delivers an average PUE of 1.22

20.  First Mega Datacentre built outside the
US, on-line on July 2009
 Cost $500 million, 303+sq ft facility and
growing
 EU Datacenter Best Practice Award Winner
 Stand-alone Server Pods
 Environmentally Sustainable
 Free Air-Cooling via Air-Side
Economization
 50% less energy use vs. traditional
facilities
 1 % water use vs. traditional facilities
 5.4 MegaWatts today, 22.2 MW for future use
 ISO 27001:2005 Accreditation
 SAS 70 Type I and II Attestations

21. Windows Azure

23. The Windows Azure Platform
The Big Picture
Windows Azure
AppFabric
Applications
SQL Azure
Windows Azure
Applications
Server Desktop Mobile Others

24. Application
Compute Storage
Config Fabric
AppFabric
Applications
SQL Azure
Windows Azure
Applications
Server Desktop Mobile Others

25. Azure Building Blocks

26. Azure Compute Building Blocks

27. Windows Azure Roles
 Your role class typically derives from
RoleEntryPoint represents role
 Do initialization in OnStart()
 Do work in Run()
 You call into RoleEnvironment to enumerate
roles, endpoints, etc
 You get called in response to changes in the
number of instances in the service
 Can force restart

28. Windows Azure Storage Building Blocks
Each storage account can hold up to 100 TB

29. Windows Azure Blobs
 Provide simple interface for storing named files
along with metadata for the file

30. Windows Azure Blob features
 REST based API
 ouraccount.blob.core.windows.net/thecontainer/
myblob
 PutBlob, GetBlob, DeleteBlob, CopyBlob,
SnapshotBlob, LeaseBlob, etc
 Block Blob for streaming + commit-based
writes
 Page Blob for random read/write
 100 Tb per storage account

31. Windows Azure Drives
 Page Blob formatted as a NTFS single volume
Virtual Hard Drive (VHD)
 NTFS APIs providing durable file storage on Page
Blobs
 Drives can be up to 1TB, up to 16 can be mounted
 Can upload VHD via Page Blob then mount
 Ease migration of existing Windows
applications to Azure
 Durability and survival of data on application
failover or hardware failure

32. Windows Azure Tables
 Structured storage via entities, containing sets of
properties

33. Windows Azure Tables
 Provides Massively Scalable Structured Storage
 Billions of entities (rows) and TBs of data
 Can utilise thousands of servers as it scales up
 Triple replicated for durability
 Familiar and easy to use REST based API
 Each entity can have up to 255 properties
 PartitionKey & RowKey as key and index
 Timestamp for optimistic concurrency
 No fixed schema for all other properties
 Property is stored as a <name, typed value> pair

34. Windows Azure Tables
MyTable1
Entity: Sports Car
Entity: Car
Entity: Customer

35. Table Partitions and Rows
PartitionKey RowKey Timestamp ReleaseDate ………..
(Category) (Title)
Action Clash of the Titans … 2010
Action The Bourne … 2007
Ultimatum
… … … …
Animation Avatar … 2009
Animation How to Train Your … 2010
Dragon
… … … …
Comedy What Happens in … 2008
Vegas
… … … …
SciFi Alien 1979
… … … …
Horror Halloween … 1978

36. Windows Azure Queues
 Provide reliable storage and delivery of messages

37. Windows Azure Queues
 Provide reliable storage and delivery of
messages
 Max size of message 8 Kb
 Store URI to blog / table for larger data
 No limit to number of messages in a queue
 Messages must explicitly be deleted
 Message expiration timeout
 Message de-queue count for poison messages

38. Common Azure Pattern

39. Azure Features

40. Service Management API
 ProvideLights-Out Service Management
 Manage services programmatically via
REST-based API
 X509 client certificates for authentication
 Viewing, creating, deleting, swapping,
modifying configuration settings, etc on
deployments

41. Windows Azure Diagnostics
 Cloud is much harder than single server
 Dynamic environment, no local access
 Azure Diagnostics built for monitoring & data
collection
 Focus on what to collect and when
 Designed for Azure
Logging &
 Based on standard APIs Diagnostics
 Traces, logs, crash dumps, IIS logs,
perf counters, ETW, etc
 Upload to Azure storage as required
41

42. Service Management CmdLets
 PowerShell cmdlets wrapping Azure Service
Management and Diagnostics API
 Simple to script out deployments, upgrades,
scaling
• Deploy new services
• Upgrade services
• Manage storage
accounts
• Transfer diagnostics
information

43. Service Upgrade Models
 Rolling upgrade (aka “In-place”)
 Roles are updated across update domains
one at a time
 Service remains available during upgrade
 Real-time hot swap (aka “VIP Swap”)
 Virtual IP swap between staging &
production
 Swap service endpoints of old/new
version
 Enables complex architectural changes
 Planned downtime
 Stop and replace service with new version

44. Content Delivery Network
 Better performance and user experience by caching
Azure blobs at strategically placed locations
 19 locations globally (United States, Europe, Asia,
Australia and South America) and growing
Edge Location Edge Location
pic1.jpg
Edge Location
Content Delivery Network
pic1.jpg
Windows Azure
Blob Service

45. The Fabric

46. Fault Domains

47. Azure Development

48. Windows Azure for Application Developers
Code
Software Development Kit Portal
Data
Relational
Data
API
Develop Deploy Run

49. Development Fabric and Storage

50. Visual Studio 2010

51. Visual Studio 2010

52. Visual Studio 2010

53. Azure Portal

54. SQL Azure

56.  account servers

Account

 server


Server

 database
Datab 
ase 

57. Application
Applications use standard SQL
client libraries: ODBC, ADO.Net, …
Internet
Azure Cloud
TDS (tcp)
Load balancer forwards ‘sticky’
LB sessions to TDS protocol tier
TDS (tcp)
Gatewa Gatewa Gatewa Gatewa Gatewa Gatewa
y y y y y y
Gateway: TDS protocol gateway, enforces AUTHN/AUTHZ policy; proxy to backend
TDS (tcp) SQL
SQL SQL SQL SQL SQL SQL
Scalability and Availability: Fabric, Failover, Replication, and Load balancing

58. SQL Azure & SQL Server
 Similarities
 “Just change the connection string”
 Same SQL Server core database service &
TDS
 Differences
 Supports subset of SQL Server 2008
 Limited to 5Gb or 50Gb database
 Different pricing model
 No Reporting Services, Analysis Services, etc

59. 







60. SQL Azure Architectures
SQL Azure access from SQL Azure access from
within the cloud outside the cloud
Application/ Browser Application / Tools
SOAP/REST
HTTP/S
Web Role
Windows Azure
T-SQL (TDS)
T-SQL (TDS)
SQL Azure SQL Azure
“Code Near” “Code Far”

61. On-Premises
Applications SQL
Azure
Sync
Group
SQL SQL
Azure Azure
Remote Offices
& Retail Stores Offline-Capable
Cached Mode Applications

63. SQL Azure Roadmap
 Service Update 2
 Multiple Active Result Sets (MARS)
 Service Update 3
 50 Gb Databases, Spatial support
 Incubation
 Project “Houston” – new SQL Azure web administration tool
 Codename “Dallas” – information as a service
 OData service on top of SQL Azure

64. 







65. OData
“ODBC for the Internet”
• Uniform way of representing data, URL conventions and
operations
• Auto generation of OData for SQL Azure
OData Consumers
Session
Reports Dashboard Forms Apps Browser
Open Data
Protocol
Format Query&[Uri] R/W Batching,Versions
Metadata
+ + +
OData Producer

66. Codename “Dallas”

67. I
need
data!
Marketplace
Content Providers
Office,
Dynamics,
Services for
SQL Server
Standard Data
Access (OData)
ISV Apps
Custom
Apps Billing
Analytics

68. Information as a Service
 Data or functionality that is “of value to many” –
enabling applications, reports, BI analysis, etc…
 Examples include GIS/Spatial, traffic, movie show times,
crime, real-estate sales, financial data, navigation, census
data, reviews, etc…
 Various Classifications:
 Commercial: clean, supported, and regularly updated from
ISVs and Content Providers
 Trusted Public Domain: clean, unsupported data from
academia and governments (Census, FDA, …)
 Crowd Sourced: unreliable data in the public domain from
anyone and everyone
68

69. http://apps.facebook.com/crimestats http://silverlight.onterrasys.com/CrimeDemo/

70. View results in
A quick way to explore “Dallas” datasets Learn about data
various formats
format and perform
quick queries
Available for
every dataset
Fill in parameter
Click
values using
“Analyze” to
guided help
continue
exploring in
PowerPivot
and Excel to
create
PivotTables,
graphs, etc.

72. AppFabric

73. “Infrastructure to help build and
manage applications more easily”
 Windows Server AppFabric
 Caching capabilities (“Velocity”)
 Workflow + service hosting (“Dublin”)
 Windows Azure Platform AppFabric
 Service Bus
 Access Control

75. Service Bus
 Exposing internal applications on the
Internet isn’t easy
 Network address translation (NAT) and
firewalls get in the way
 The Service Bus:
 Provides a cloud-based intermediary
between clients and internal applications
 Provides a service registry that clients can
use to find the services they need
75

76. Service
Bus
Send Receive Receive Send
App #1: Web, App #2: Web,
Mobile, Desktop, App 1 App 2 Mobile,
LOB, RIA, ... Desktop, LOB,
RIA, ...
without altering network
topology

77. Access Control
 Different organizations identify users
through many different techniques
 Applications can be faced with a confusing
mess of security related code
 The Access Control Service:
 Implements a security token service (STS) in the
cloud
 It accepts one token and issues another
 An administrator can define rules for how this
claims transformation is done
77

78. • Many identity providers, one code base

79. 








March ‘10 V-Next
Labs  Advanced messaging features (ie. durability)
 Silverlight support  BizTalk Server integration
 Multicast with Message  Windows Server AppFabric integration
Buffers  Web SSO and service federation - WS-Federation
 Considering additional identity providers
 Windows Live ID, Open ID, Facebook Connect
 Management UI

80. Commercial Information

82. Windows Azure Platform Consumption Prices
$0.12/hour $0.15 GB/month $9.99/month $99.99/month
+ Variable Instance Sizes $0.01/10K transactions (scale up to 5 GB) (scale up to 50 GB)
Per Message Operation Per Connection
$1.99/100k Message Operations $3.99/connection
Prices shown in USD only International prices are available

83. Windows Azure Instance Sizes
$0.12 $0.24 $0.48 $0.96
Unit of Compute Defined
1 x 1.6Ghz 2 x 1.6Ghz 4 x 1.6Ghz 8 x 1.6Ghz
1.75 GB memory 3.5 GB memory 7.0 GB memory 14 GB memory
250 GB storage 500 GB storage 1000 GB storage 2000 GB
(instance storage) (instance storage) (instance storage) (instance storage)

84. Windows Azure Platform Data Transfer
North America Region Europe Region Asia Pacific Region
N. Europe
N. Central – US Sub-region
Sub-region
W. Europe E. Asia
Sub-region Sub-region
S. Central - US
Sub-region
S.E. Asia
Sub-region
On-board to Windows Azure platform at no charge
Off peak times defined as: 10pm-6am Mon-Fri & from 10pm-Fri to 6am-Mon for
weekends in each designated regional time zones below
North America Europe Asia Pacific
PST = UTC-8 WET = UTC SST = UTC+8

85. Enabling customers to deliver content closest to end-users. Leads to better user
experience, global reach, & engagement

87. Introductory MSDN Biz Spark
Special Premium One
Anyone can try the platform Subscribers receive monthly Investing in high potential
at no charge (monthly allocation of platform services start-ups; receive full platform
service allocations) at no additional charge services at no charge
Development Accelerator

88. UK Azure Pricing – Standard Rates
 Windows Azure  AppFabric
 Compute  Access Control
 Small instance (default): £0.0728 per hour  £1.2062 per 100,000 transactions*
 Medium instance: £0.1455 per hour  Service Bus
 Large instance: £0.291 per hour
 £2.4184 per connection on a “pay-as-you-go”
 Extra large instance: £0.5819 per hour basis*
 Storage  Pack of 5 connections £6.0307*
 £0.091 per GB stored per month  Pack of 25 connections £30.1535*
 £0.0061 per 10,000 storage transactions
 Pack of 100 connections £120.6139*
 Content Delivery Network (CDN)
 Pack of 500 connections £603.0695*
£0.09 GB CDN Egress
£100 per month* - Highly Available 2 x

 Data Transfers
£0.06/100K CDN Storage X-actions 
 SQL Azure  North America and Europe regions

InstanceUp to 1 GB relational
Web Edition –
database Web Application + 0.5TB per GB out
£0.091
data
£0.0607 per GB in 

 £6.055 per database per month  Asia Pacific Region
 Business Edition – Up to 10 GB  £0.1819 per GB in
relational database
£6 per month* - Highly Available 1Gb
 £60.604 per database per month

 £0.2728 per GB out
Inbound data transfers during off-peak times
through June 30, 2010 are at no charge. Prices
SQL Server database revert to our normal inbound data transfer rates
after June 30, 2010
* Lots of assumptions and caveats

89. Some Final Thoughts….

90. © 2010 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.