Windows Azure Realized

27 

Full text

(1)

© Copyright 2009 Avanade Inc. All Rights Reserved.

The Avanade name and logo are registered trademarks in the US and other countries.

Windows Azure Realized

Steffen Vorein

Lead Technologist, Cloud Strategy& Windows Azure

Avanade Inc, Global Technologies and Solutions

(2)
(3)

© Copyright 2009 Avanade Inc. All Rights Reserved.

Challenges

• Perceived Risk of New Platform

• Lack of Developer Expertise in Tools and Patterns

• Compute Elasticity & Load Management

• Identity & Access Management

• Performance Monitoring and Management

• Data Security Concerns and Regulations

(4)
(5)

© Copyright 2009 Avanade Inc. All Rights Reserved.

Batch Style Architecture

(6)

Elasticity

• Runtime cost management becomes part of the architecture

• No “instant-on” for new instances

– Can take up to 60’ to spin-up new instances

– Spin-up instances in advance if burst expected

 Could be on schedule (every Monday morning 8pm, etc.)

• No shutdown of a specific instance

– Makes downscaling difficult with long running processes

• Role instances are billed per partial wall clock hour

– Keep instances alive until shortly before a full hour ends

– Keep a spare instance running

• 2+ instances required for SLA

– Instances are distributed in upgrade groups which get upgraded

sequentially

(7)

© Copyright 2009 Avanade Inc. All Rights Reserved.

Load Distribution

• Variable number of compute instances

– Unlike a grid where fixed set of resources is available

• Key factors for a batch style application

– Queue Length

 Number of items in queue

 Estimated workload (per item)

– Average throughput per instance

– Maximum allowed processing time

– Minimum/Maximum number of instances

(8)

Diagnostics: Many Data Points

• Assume sampling 10 performance counters every 30 seconds on an

Azure application with 2 Web Roles and 2 Worker Roles. Each Web

Role has 2 instances and each Worker Role has 2 instances:

• 8 total instances to track

• 80 records every 30 seconds, or 160 records per minute

230,000 Data Points Per Day

• Search Engine: Twitter Feed Search Indexer

• ~100 Instances, Tracking 9 performance counters, 30 second

sampling rate

• 900 records per 30 seconds = 1,800 records per minute

• 1,440 minutes per day * 1,800 records per minute =

(9)

© Copyright 2009 Avanade Inc. All Rights Reserved. 9

Customer Scenarios

Realized on Azure

(10)
(11)

© Copyright 2009 Avanade Inc. All Rights Reserved.

eBay

eBay is the world’s largest marketplace and is a world leader in shopping and payments in the web

11

Azure Solution

Developer

Eclipse IDE

Users

Compute Nodes Running eBay Apps (Tomcat) Compute Nodes running Database (MySQL) Provisioning Node Azure Storage Request /Status Data (MySQL connection) Request/ Status Logs Logs

Original Solution

(12)

eBay

(13)
(14)

Emergency Web Solution

Normal Operation

 A Publish Area (content staging) exists where

content can be prepared before it is visible to

the public

 Publishing area can accept content both via

manual updates and via XML feeds

Runtime

Normal

Publish

(content staging) Normal

XML

Content Editor

!

Emergency Web

Publish

(content editing) Normal

Runtime

Normal

Emergency Mode

Customer can themselves trigger “Emergency

Mode”

 Emergency Web is a simplified high

performing web site and is separated from

normal web to isolate capacity issues under

extreme load

 Emergency Web can be deployed to data

center closest to the event region

(15)
(16)

Azure Solution

Origin Digital Servers

Users

Transcoding

Engine

CDNs

Digital Content

Billing

Users

Digital Content

Transcoders

Origin Digital Servers

Billing

Transcoder

Manager

CDNs

(17)
(18)

Grid Architecture Framework

The Grid Concept leverages the availability of computing resources for multiple/parallel task execution

Original Solution

(19)
(20)

Azure Solution

Enterprise Security Logging Performance Error Management Remoting Reference Tables Channel/Web State Management

Navigation Text/String I18N

Validation Configuration Currency

Data Access Asynchronous ManagementIdentity

Transaction Management File Management MQ Series Custom Web Controls Custom Data Type Custom Data Type PC Interface Process Controller PC Creator SC Interface SC Creator Server Component

FNOL Presentation

FNOL Business

FNOL Data

AICS Presentation

AICS Business

AICS Data

On-Premises Servers

AICS DB

Claims : Existing Logical View

Windows ServicesWindows ServicesWindows

Services UI Context

ASPX

Code

Behind

ASPX

Code

Behind

ASPX

Code

Behind

BG Proc

Original Solution

(21)
(22)

Azure Solution

On-premise

Dealer

SQL Server

Dealer

IIS 6.0

Cobol.Net

Original Solution

SQL Azure

IIS 7.0

Cobol.Net

MicroFocus

MicroFocus

(23)
(24)

Archive

(.NET)

Service engine

(.NET, WCF,

Forms Server)

Statistics and

analysis

(Para Data)

(SQL, SSIS)

Service owner systems Public registers MyID

Authentication

(3rd Party OpenSSO, JEE)

3rdparty services

Authorization

Other Web sites

Intermediary

(BizTalk)

Payment

(.NET, WCF)

Payment provider

My

Profile

Certificate Authority End User systems BuyPass BankID Enterprise certificate Agency systems

Register

(.NET, WCF)

End User/ Agency User

Portal

(SharePoint)

(25)

© Copyright 2009 Avanade Inc. All Rights Reserved. 25

Archive

(.NET)

Service engine

(.NET, WCF,

Forms Server)

Statistics and

analysis

(Para Data)

(SQL, SSIS)

Service owner systems Public registers MyID

Authentication

(3rd Party OpenSSO, JEE)

3rdparty services

Authorization

Other Web sites

Intermediary

(BizTalk)

Payment

(.NET, WCF)

Payment provider

My

Profile

Certificate Authority End User systems BuyPass BankID Enterprise certificate Agency systems

Register

(.NET, WCF)

End User/ Agency User

Portal

(SharePoint)

(26)

http://azure.com

http://wag.codeplex.com

– Patterns&Practices guidance and links

http://thecloudadvantage.com/

(27)

© Copyright 2009 Avanade Inc. All Rights Reserved.

Thank you

Figure

Updating...

Related subjects :