Amazon Web Services Migration Scenarios: Web Application Architecture

October 2010

Migration Scenario: Migrating Web Applications to the AWS Cloud

Figure 1: CompanyACloud.com Architecture (Before and After Migration)

Use case
CompanyACloud.com is a customer-facing web application of company A, which serves as a marketing portal and a
customer management system. Customers, partners and employees use the web application to collaborate with each
other using a rich web interface that can be viewed in a standard internet browser. CompanyACloud.com lists a
complete catalog of products and their details. As new product announcements are made, marketing campaigns
generate substantial amounts of traffic to the site resulting in periodic spikes. Outside of the timeframes caused by
these spikes, CompanyACloud.com experiences a fairly steady and predictable traffic load, which is characteristically
high on weekdays and low on weekends. The website is currently hosted on dedicated infrastructure at the companys
headquarters.

Application Architecture
Using a standard 3-tier application architecture, the company deploys a frontend hardware-based load balancer, which
manages traffic across two Apache web servers each running on a separate physical box. The application is running

Page 1 of 4

Amazon Web Services Migration Scenarios: Web Application Architecture

October 2010

behind the company firewall (DMZ) and uses standard SSL encryption. The backend business logic is implemented in
Java, and leverages Tomcat as application container and application server, and three Tomcat servers power the
website. The application also had a database layer with consists of one master MySQL server and two slave servers for
greater performance. A simple illustration of the architecture is provided in Figure 1 on the left hand side.

Motivation for Migration

Company A would like to move the web app to the cloud environment for three main reasons.
1. The company wishes to scale out the web application, and address the growing traffic, without investing in new
hardware.
2. They would like to cut down administration costs by automating deployments.
3. Finally, the company would like to expand and provision extra capacity only when its needed, for example,
when running marketing campaigns.

Migration Plan, Strategy, & Execution Steps

Cloud Assessment
On financial assessment, the technical program manager from the website development team was able to map the
hardware configuration of physical servers to equivalent EC2 instance types and estimate the combined storage and
bandwidth requirements. The team realized that they could free up the current infrastructure for other internal
projects, discontinue a tape backup maintenance contract and reduce their operating expenditures by 30%.
During the technical assessment, they discovered that the entire CompanyACloud.com technology stack was compatible
with AWS and could run on Amazon EC2 Instances with Linux. They also discovered that the web app can be configured
to run at peak capacity (7 Servers) during promotional campaigns, medium capacity (4 Servers) on weekdays and low
capacity (2 Servers) on weekends.
The IT Security team was able to get a complete SAS 70 Type II audit report from AWS and they were able to review
security best practices1.

Proof of Concept
The web development team was skeptical about the relational database migration. To test, they decided to build a proof
of concept application. During the proof of concept, the team learned the following techniques: starting, terminating
and configuring Amazon EC2 instances and Amazon RDS DB Instances, storing and retrieving Amazon S3 objects, and
setting up elastic load balancers using the AWS Management Console2. They learned a ton about AWS and saw that
they have full control over the environment, and felt a lot more confident about moving to the next step. The relational
database files (binary and transaction logs) were moved to Amazon RDS instances using the standard mysqlimport
utility. For the test environment, they deployed a DB Instance within a single Availability Zone, and for the production
environment, they set up a DB Instance with the Multi-AZ deployment to increase availability. The team was able to
successfully test and migrate all data to a DB instance, get performance metrics using Amazon CloudWatch, and set

1
2

AWS Security Center http://aws.amazon.com/security

AWS Management Console http://console.aws.amazon.com

Page 2 of 4

Amazon Web Services Migration Scenarios: Web Application Architecture

October 2010

retention policies for backups. They built migration scripts to automate the process and created awareness within the
organization by organizing a brownbag session and successfully demonstrated their work to their peers.

Data Migration
Once the proof of concept was complete, the team decided to move all of the applications static files (Images, JS, CSS,
video, audio, and static HTML content) into an Amazon S3 bucket, created a CloudFront distribution of that Amazon S3
bucket, and modified the references in web pages so that end-users get the content directly from Amazon S3 and
Amazon CloudFront. With a few scripts and the AWS SDK for Java library, they were able to transfer all data from tape
drives and upload it to Amazon S3.

Application Migration
During the application migration phase, the development team launched both small and large instances3 for their web
and tomcat servers. They created AMIs (Amazon Machine Images, basically golden system images) for each server
type. AMIs were designed to boot directly from an EBS volume and fetch the latest WAR file binaries during launch from
the source code repository. They modified their build and deployment scripts to use the cloud as an endpoint. Security
Groups were defined to isolate web servers from the applications and database servers. Testing (functional, load,
performance etc.) was performed to ensure that the systems were performing at expected levels, and that exit criteria
for each component were met.

Co-existence Phase
During the migration phase, the collocation infrastructure
was not deprecated immediately. Company A employed a
hybrid migration strategy during the migration of all web
and application servers. The configuration of the onpremise hardware load balancer was modified to send
requests to the new instances in the cloud. For a short
duration, the load balancer was routing traffic to the servers
in the cloud in as well as to the physical servers. After
verifying that the servers in the cloud were performing at
required levels, the physical servers were dismissed one by
one, the load balancers were updated, and all of the web
traffic was being served up by the EC2 instances running in
the cloud.
After testing was completed, the DNS was switched to point to the cloud-based web servers and the application was
fully migrated to the AWS cloud.

Leveraging the Cloud

Once the production site was launched, Company A was looking forward to the time when they could use some of the
advanced features of AWS. The team automated some processes so that once the server is started it could be easily
attached to the topology. They created an Auto Scaling group of web servers and were able to provision more
capacity automatically when specific resources reach a certain threshold (Apache web servers CPU utilization above 80%
for 10 min). The team invested some time and resources in streamlining their development and testing processes to

Amazon EC2 Instance Types http://aws.amazon.com/ec2/instance-types

Page 3 of 4

Amazon Web Services Migration Scenarios: Web Application Architecture

October 2010

make it is easy to clone testing environments. They gained lot of experience using AWS resources and also invested time
in leveraging multiple Availability Zones for even higher availability.

Optimization
During the optimization phase, the development team analyzed their utilization patterns and realized that they could
save 30% if they switched to Reserved Instances4. They purchased
four Reserved Instances (2 for web servers and the other 2 for tomcat
Mode
Web Servers App Servers
servers). They built additional scripts to run their web application in 3
Weekend
1
2
different modes: weekend, weekday and promotion mode. These
Weekday
2
3
modes defined the minimum number of servers to run. The team also
Promotion
5
7
integrated Amazon CloudWatch into their existing dashboards so that
they can monitor the system metrics of every instance in their cloud fleet.

Conclusion
The company was able to successfully migrate an existing web application to the AWS cloud. With minimal effort, the
team was not only able to free up the physical infrastructure for other projects but also reduce the operating
expenditure by 30%. Using the phased-driven approach, the development team was able to resolve all the financial,
technical and social-political concerns. Deciding to invest in a proof of concept proved extremely valuable. The resulting
architecture was not only elastic and scalable but also flexible and easier to maintain.

Amazon Reserved Instances http://aws.amazon.com/ec2/reserved-instances

Page 4 of 4