In this article i would like to share some of the insights on AWS Auto Scaling in following perspectives: • Need for Auto Scaling • How AWS Auto scaling can help to handle the various load volatility scenarios • How to configure an Auto scaling policy in AWS • Things to remember before Scaling out and down • Understand the intricacies while integrating Auto scaling with other Amazon Web Services • Risks involved in AWS Auto scaling

1. Auto Scaling using AWS
Harish Ganesan
Co founder & CTO
Contact: Harish11g.AWS@gmail.com
www.twitter.com/harish11g
http://www.linkedin.com/in/harishganesan

2. Introduction
• Need for Auto Scaling
• How AWS Auto scaling can help us ?
• How to configure an Auto scaling policy in
AWS
• Things to remember before Scaling out and
down
• Understand the Intricacies while integrating
Auto scaling with other Amazon Web Services
• Risks involved in AWS Auto Scaling

3. Key AWS terms used
EC2 : Elastic Compute Cloud
EBS : Elastic Block Storage
S3 : Simple Storage Service
SQS : Simple Queue Service
ELB : Elastic Load Balancing
AMI : Amazon Machine Images
AZ : Amazon Availability Zones ( Data centers)
VPC : Amazon Virtual Private Cloud
RDS : Amazon Relational Database Service

4. Key Scaling terms
Scale out : Achieving scalability by increasing the
number of Amazon EC2 instances
Scale up : Achieving scalability by resizing the
capacity( Compute, Memory and EBS) of existing
Amazon EC2 instances
Scale down : Decreasing the number of Amazon
EC2 instances or the configuration for existing EC2
instances

5. Need for Auto Scaling

6. Upfront Capacity Investment
Source: AWS

7. Actual Demand vs Fixed Capacity
Source: AWS

8. Problem 1: Wasted Capacity
Source: AWS

9. Problem 2 : Lost Customers
Source: AWS

10. Need for Auto Scaling
Source: Internet

11. What is AWS Auto Scaling ?
AWS Auto Scaling allows us to scale our Amazon
EC2 capacity out or down automatically according
to the load patterns . Example :
• We can expand the number of Amazon EC2 instances from 1 to
100+ automatically during load peaks
• We can reduce the number of Amazon EC2 instances from 100+ to
1 automatically during load valleys

12. AWS Auto Scaling
We can closely align our Infrastructure
with our load requirements and save
costs
Source: AWS

13. Load Volatility Scenarios

14. Candidates for Auto Scaling (1)
1 Predictable Bursts • Applications which has
Compute
following Load Volatility
characteristics like
– Predictable bursts of request
load
Time
– A Pattern in Spikes and lows
– “On” and “Off” loads
– Seasonal trends

15. Candidates for Auto Scaling (2)
2 Un Predictable Bursts • Applications which has
Compute
following Load Volatility
characteristics like
– Un Predictable bursts of
request load
Time
– No specific pattern in Spikes
and lows
– Highly elastic ,unexpected
and scalable demands

16. Candidates for Auto Scaling (3)
3 Constantly Growing • Applications which has
Compute
following Load Volatility
characteristics like
– Constantly growing
– Successful and growing very
Time
fast
– Complex lead time for
deployment

17. What Amazon Auto Scaling can do ?
• Handle all the 3 load scenarios (Candidates)
• Scale out Amazon EC2 instances seamlessly and
automatically when demand increases
• Scale down unwanted Amazon EC2 instances
automatically and save money when demand
subsides
• Decide the scaling based on AWS CloudWatch
metrics
• Auto Scale your Web servers(Amazon EC2) in
combination with AWS Elastic Load Balancing

18. How much does it Cost ?
Cost for using AWS Auto Scaling service = 0$
Value= PRICELESS
Note : AWS Auto scaling needs Amazon CloudWatch
monitoring service to function . Amazon
CloudWatch is billed on usage basis.

19. Some AWS Auto scaling Concepts
Auto Scaling group : Logical grouping of multiple
Amazon EC2 instances for easy scaling and
Management
Health Check: Calls to check on the health status of
each Amazon EC2 instance in an Auto Scaling group
Launch Configuration: Captures the parameters
necessary to create new EC2 instances in Auto Scaling
mode

20. Some AWS Auto scaling Concepts
Triggers: A CloudWatch alarm and an Auto Scaling
policy that describes the actions when the alarm
threshold is crossed . Two Triggers – Scaling out and
Scaling down needs to be created
Policy : Set of instructions for Auto Scaling that tells
the service how to respond to AWS CloudWatch alarm
messages

21. Amazon AutoScaling Architecture
EC2 Instance
Elastic Load
Balancer
… Amazon CloudWatch
Scale Up
Key Concepts : Scale Up
Rule
EC2 Instance
• Launch Config Scale Down
Scale
Down
• Auto Scaling Group Rule
• Trigger Scale Up
• Policy Predefined
Scaling
• Alarm Auto Scaling Scale Down Activity
Group

22. Notable parameters on Amazon Auto Scaling
• The percentage of allocated EC2 compute units that are
CPUUtilization currently in use on the instance
• Filters and responds with data that we request about all
AutoScalingGroupName the Amazon EC2 instance part of the Auto scaling group
• Time taken between a request and the corresponding
Latency response as seen by the AWS Elastic Load Balancer
• The number (sum) of requests processed by the AWS
RequestCount Elastic Load Balancer
• The number (average) of healthy instances behind the
HealthyHostCount AWS Elastic Load balancer
• The number (average) of unhealthy instances behind the
unHealthyHostCount AWS Elastic Load Balancer

23. Steps to configure Amazon Auto Scaling with
AWS Elastic Load Balancing

24. Step 1: Configuring AWS Auto Scaling with
AWS ELB
elb-create-lb my-load-balancer --headers --listener "lb-port=80,instance-
port=8080,protocol=HTTP" --availability-zones us-east-1c
App server port to Add a name to your The load
which requests needs load balancer balancer port
to be forwarded

25. Step 2: Create a launch configuration
as-create-launch-config my-lconfig --image-id ami-e3826c8a --instance-type m1.small --key
my-key-pair --group my-security-group
Key pair / Security Name your launch Amazon Machine Amazon EC2
group settings for the configuration image(AMI) to be Instance Size
Amazon EC2 instances launched during
scaling

26. Step 3: Create an AWS Auto Scale Group
as-create-auto-scaling-group my-as-group --availability-zones us-east-1c
–launch-configuration my-lconfig --max-size 11 --min-size 3 --cooldown 180 --load-
balancers my-load-balancer
Load balancer name in Name your Auto Availability Zone in Minimum/Maximum
which the new Amazon scale group which the Auto number of Amazon
EC2 instances launched scaled Amazon EC2 EC2 instances
will be attached instances will be maintained by Auto
launched Scale

27. Step 4: Configure the Auto scaling Triggers
as-create-or-update-trigger my-as-trigger --auto-scaling-group my-as-group --namespace
"AWS/EC2" --measure CPUUtilization --statistic Average --dimensions
"AutoScalingGroupName= my-as-group " --period 60 --lower-threshold 20 --upper-threshold
80 --lower-breach-increment"=-2" --upper-breach-increment 4 --breach-duration 180
Measure the Scale out by 4 Lower CPU Limit is
average CPU of Amazon EC2 20% and Upper CPU
the Auto Scale instances. Limit is 80%
Group Scale down by 2
Amazon EC2
instances

28. Applying Amazon Auto Scaling for 3 different
load scenarios

29. Auto Scaling Policy : Predictable Bursts
1 Predictable Bursts
• Configure Time based Auto
Scaling plan
Compute
• PutScheduledUpdateGroup
Action - specify the date
and time of execution,
Time minimum, maximum, and
desired Instance size of
capacity
% as-put-scheduled-update-group-action my-group –name “My-scale-
up” -- time “2011-04-05T02:00:00Z” -min 10 -max 50

30. Auto Scaling Policy : Un predictable Bursts
2 Un Predictable Bursts • Configure On demand Auto
Compute
Scaling policy
• Create policies for both
Scaling out and Scaling
down scenarios
Time
• Auto Scaling will respond to
changing conditions
dynamically
%as-create-auto-scaling-group as-group --availability-zones us-east-1c –
launch-configuration my-lconfig --max-size 20 --min-size 2 --cooldown
180 --load-balancers my-load-balancer

31. Auto Scaling Policy : Constantly Growing
3 Constantly Growing • Periodically monitor the
Compute
load requirements and
manually configure the
Auto scaling parameters
• Increase the minimum fixed
Time
capacity gradually
• Leverage On-Demand
Amazon EC2 instances
%as-create-auto-scaling-group my-as-group --availability-zones us-east-
1a –launch-configuration my-lconfig --max-size 4 --min-size 2 --
cooldown 360 --load-balancers my-load-balancer

32. Important points to remember before Scaling
out / down using AWS Auto scaling

33. While Scaling out
• Very frequent (minutes) scaling out activity is not
cost effective
• Scaling out with smaller instance type units closely
aligns our capacity with our load requirements
• We can create a maximum of 125 scheduled actions
per Auto Scaling group. This allows scaling four
times a day for a 31-day month for each Auto
Scaling group
• We can configure a scheduled Scaling action for up
to a month in the future

34. While Scaling down
• Amazon Auto scaling does not close the Amazon
EC2 instances in the same of order of launch
• To avoid data inconsistency architects should ensure
the following points on Amazon EC2 servers of
Auto scale group
• Log files generated by Auto Scaled Amazon EC2 are
moved to Amazon S3 periodically
• Data files uploaded in the Web/App EC2 instances are
moved to S3 or GlusterFS
• Sessions of Web/App layer and Application Cache layers
are replicated to a centralized /distributed resource

35. Candidates for Auto Scaling
• Web Tier ( Apache , Nginx etc)
• Application Tier ( Tomcat , Jboss etc)
• Load Balancing Tier ( HAProxy, Nginx etc)
• Other Stateless Tiers

36. Should not use Amazon Auto Scaling for?
• Database Tier ( MySQL , SQL , Oracle etc) automate
RDS Read Replica’s or Read Slaves Scaling out using
custom Scripts
• Use Custom Scaling programs / Scripts to add
new nodes into the replications or clusters or
Shards in the following
• Distributed Caching Tier ( memCached )
• Solr Search engine
• NoSQL ( Redis , MongoDB) .

37. Tips ,Pointers and Notes:
While integrating Amazon Auto Scaling with
other AWS components

38. Amazon Auto scaling and Amazon Machine
Images (AMI’s)

39. Amazon Auto Scaling and AMI’s
• Amazon Auto Scaling can launch new EC2 instances
from S3 backed AMI’s and EBS backed AMI’s
• AMI’s should be present on the Same Region where
the EC2 instances will be launched in Auto scale
group
• AMI’s need not be present on the Same AZ where
the EC2 instances will be launched in Auto scale
group

40. S3 Backed AMI’s for Auto Scaling
1 1 Web requests are sent to the
Amazon ELB
Amazon Elastic Load Balancer
2 AWS ELB transfers the requests to
2 AZ : us-east-1a Amazon EC2 instances launched in
us-east-1A availability zone
AWS Security Groups
3 Amazon EC2 instances are
configured to be part of AWS Auto
Scale group
3
EC2 Instances 4 AWS Auto Scaling launches the new
Amazon EC2 instances from the S3
backed AMI’s
Amazon Auto Scaling
4

41. EBS Backed AMI’s for Auto Scaling
1 1 Web requests are sent to the
Amazon ELB
Amazon Elastic Load Balancer
2 AWS ELB transfers the requests to
2 AZ : us-east-1b Amazon EC2 instances launched in
us-east-1B availability zone
AWS Security Groups
3 Amazon EC2 instances are
configured to be part of AWS Auto
Scale group
3
EC2 Instances 4 AWS Auto Scaling launches the new
Amazon EC2 instances from the EBS
backed AMI’s
EBS
EBS
EBS
Amazon Auto Scaling
4

42. Amazon Auto Scaling and AMI’s
• EBS backed AMI’s are faster to launch compared to
S3 backed AMI’s
• EBS backed AMI’s (Windows) launch in ~20 seconds
(approx)
• S3 backed AMI’s (Windows) launch in 3 minutes (approx)
• Sometimes it is better to have EC2 instances
launched faster in Auto scaling when overall CPU
threshold is breached
• EBS backed AMI’s option in AWS Auto scaling may
not be cost effective compared to S3 backed AMI’s

43. AWS Auto scaling and Amazon EC2 instances

44. Amazon Auto Scaling and Amazon EC2
• A single Amazon Auto Scaling group can launch only
one type of EC2 instances
• Example , We cannot configure Amazon Auto scaling
group to launch combinations of both m1.small and
m1.large instance types
• A single Amazon Auto Scaling group can launch only
one version of AMI
• Example , We cannot configure Amazon Auto scaling
group to launch 2 versions of AMI bundled

45. Amazon Auto Scaling and Amazon EC2
• We can combine Reserved Instances (RI) and On-
Demand EC2 Instances in our Scaling architecture
• Amazon Auto scaling cannot be used to scale
Amazon RDS instances

46. Amazon Auto scaling and AWS regions/AZ’s

47. Design 1: Amazon Auto Scaling across AZ’s
HTTP requests are redirected to Amazon EC2 instances
configured across the AWS availability zones in US East
Amazon Elastic Load balancer
AZ: US-EAST-1A AZ:US-EAST-1B
Amazon Security Groups Amazon Security Groups
C C
L L
Amazon EC2
O Amazon EC2
O
instances U instances U
D D
W W
Amazon Auto Scaling A Amazon Auto Scaling A
T T
S3
S3 C Backed
C
Backed H AMI H
AMI

48. Design 2: Amazon Auto Scaling across AWS Regions
Architecture Not Possible
Amazon Elastic Load balancer
AWS USA East Region AWS Europe Region
Amazon Security Groups Amazon Security Groups
C C
L L
Amazon EC2
O Amazon EC2
O
instances U instances U
D D
W W
Amazon Auto Scaling A Amazon Auto Scaling A
T T
S3
S3 C Backed
C
Backed H AMI H
AMI

49. Amazon Auto Scaling ,AWS Regions and AZ’s
• AWS Auto Scaling groups can work across multiple
AWS Availability Zones inside same Region
• If an AWS Availability Zone becomes unavailable,
Auto Scaling will automatically redistribute
applications to a different Availability Zone
• AWS Auto Scaling groups cannot work across
Regions
• Example , we cannot configure a same AWS Auto scaling
group to distribute and create Amazon EC2 instances in
both USA and Europe

50. Amazon Auto scaling and AWS Elastic Load
balancer

51. Case 1: AWS Auto Scaling with AWS Elastic Load balancing
1 1 This Architecture is suitable for Web
Layer only . AWS ELB is attached to
the AWS Auto Scaling group
Amazon Elastic Load Balancer
2 AWS ELB acts as the gateway and
2 AZ : us-east-1a transfers the HTTP requests it
receives to Amazon EC2 instances in
round robin schedule
AWS Security Groups
3 Amazon EC2 instances are
configured to be part of AWS Auto
Scale group
3
EC2 Instances 4 AWS Auto Scaling launches the new
Amazon EC2 instances
Amazon Auto Scaling
5 AWS Auto Scaling ,AWS CloudWatch
and AWS Elastic Load Balancing
works in union
4
S3 backed AMI

52. Case 2: Amazon Auto Scaling without AWS ELB
Amazon EC2 2 1 Reference Auto scaling Architecture
instances of in AWS not using AWS ELB
Auto scale group
reads messages
from AWS SQS
and processes
them SQS 2 Messages needed for processing are
put in the AWS SQS using message
producer clients
3
Master EC2 instance node
controls the Auto scaling EC2 Master EC2 Node 3 Master Amazon EC2 instance node
instance group (scale out/down) controls the scale up / scale down of
activities programmatically
the processing nodes using AWS
Auto Scaling programmatically
4
EC2 Instances 4 Auto scaled Amazon EC2 instance
processing nodes picks the messages
from Amazon SQS
Amazon Auto Scaling
Amazon EC2
instances are auto
scaled
programmatically
S3 backed AMI depending upon
SQS Load

53. Amazon Auto Scaling and Amazon ELB
• Auto Scaling supports, but does not require, AWS
Elastic Load Balancing
• We can add Elastic Load Balancing to our Auto
Scaling group and use Elastic Load Balancing metrics
(request latency or request count) to scale our
application
• During the Scaling down process, Amazon Auto
Scaling instructs the AWS load balancer to remove
the Amazon EC2 instance from the load balancing
group first

54. Amazon Auto scaling and Amazon Elastic IP

55. Amazon Auto Scaling and Amazon Elastic IP
• Every new Amazon EC2 instances launched by
Amazon Auto scaling has a Private and Public IP
address
• Configuring Elastic IP for Amazon EC2 instances
launched by Amazon Auto Scaling is not a cost
effective option
• We need to pay Amazon for unused Elastic IP’s after the
Scale down process
• Combining Elastic IP & Auto scaling should be analyzed
fully before applying in specific use cases

56. Amazon Auto scaling and Monitoring

57. Amazon Auto Scaling and Monitoring
• Many traditional infrastructure monitoring systems
rely on Agents to be installed on Amazon EC2
instances and get registered with Centralized
monitoring server for receiving alerts
• Since the Amazon EC2 servers are created and
terminated dynamically by AWS Auto scaling, it will
cause problems like
• When new EC2 instances are created by Auto scaling the
monitoring agents installed in them cannot auto register
with the Centralized monitoring server

58. Amazon Auto Scaling and Monitoring
• When EC2 instances are killed by Auto scaling , the
centralized monitoring server will misunderstand this
action as instance non-availability and creates
unnecessary noise
• Hence before integrating monitoring systems into
our AWS Auto Scaled Infrastructure we should
ensure that it can monitor dynamically spawned
Amazon EC2 instances
Note: Contact us if you want Nagios monitoring and Amazon
AutoScaling to work together

59. Amazon Auto Scaling and AWS CloudWatch
• AWS CloudWatch can monitor Auto scaled EC2
instances
• AWS CloudWatch should be configured “detailed
monitoring” option for the collection of Auto
Scaling group metrics
• Configure Auto Scaling with group metrics to send
aggregated data to Amazon CloudWatch every
minute
• Auto scaling takes appropriate Scale out /down
actions based on the alarms it receives from
CloudWatch

60. Amazon Auto Scaling and AWS CloudWatch
• Custom CloudWatch Metrics with Amazon Auto
Scaling is powerful combination
Example : Monitor Tomcat valves and send Number of
sessions input to CloudWatch, Now AutoScaling can trigger
new EC2 based on number of sessions in Tomcat.

61. Amazon Auto scaling and Launch
Configurations

62. Auto Scaling and Launch Configurations
• The maximum number of launch configurations per
AWS account is 100
• If we modify our launch configuration, AWS Auto
Scaling will not apply our new settings to existing
Amazon EC2 instances
• We can attach only one launch configuration to an
AWS Auto Scaling group at a time
• When AWS Auto Scaling wants to scale down, it first
terminates instances that have an older launch
configuration

63. Amazon Auto Scaling and Elastic Beanstalk
• Amazon Elastic Beanstalk is a deployment tool used
to automatically handle the infrastructure details
like capacity provisioning, load balancing, auto-
scaling, and application health monitoring
• We cannot leverage Elastic bean stalk based Auto
scaling in following scenarios
– If our system demands Non Java applications to be
configured on Auto scaling mode
– If our application contains Java based Background
programs (not a web app ) and they need auto scaling
facility

64. Amazon Auto scaling and Amazon VPC

65. Amazon Auto Scaling and Amazon VPC
• Amazon Auto Scaling can scale out /down Amazon
EC2 instances inside a Amazon VPC network
• AWS Elastic Load balancer cannot be tied to
Amazon Auto Scaling inside Amazon VPC – Now
possible

66. AWS Auto Scaling illustration
Following slides illustrates a visual demo of AWS
AutoScaling case
• Image 1: Amazon EC2 Servers at initial Stage
• Image 2: Load Increases : Amazon EC2 Servers are
ramped up using AWS AutoScaling
• Image 3: Load Decreases
• Image 4: Amazon EC2 Servers are dynamically
scaled down using AWS AutoScaling
The below screenshots represent a custom product developed by us using AWS
AutoScaling

67. Image 1 : Initial Stage
Screenshot represents a custom product developed by us

68. Image 2 : Load Increases

69. Image 3 : Load Decreases
Screenshot represents a custom product developed by us

70. Image 4 : Scaling down the EC2 instances

71. Risks involved in AWS Auto Scaling
Risk 1: AWS Auto Scaling takes between 30 – 180
seconds sometimes to launch a new instance(s) .
This intermediate time may cause impaired
performance for our customers
Risk 2: AWS Auto Scaling cannot differentiate
between valid (vs) malicious traffic , it can scale out
servers even for malicious traffic

72. Auto Scaling is an Art
• Auto Scaling will not work for every situation and
every application
• Just by deploying our application in AWS , does not
mean our application will leverage Auto Scaling and
scale out automatically
• To leverage Auto scaling effectively in-depth
architectural expertise is needed in both Application
and Infrastructure front

73. Auto Scaling is an Art
• In some cases Auto Scaling is an on-going(monitor
and fine tune based on trends/patterns) and not an
one time activity
• Badly configured Auto Scaling parameters will
– increase the cost of infrastructure
– cause revenue leakage rather than reducing them
– create unnecessary excess capacity

74. General Cost Savings using Auto Scaling
1 30 % savings 2 35 % savings
Compute
Compute
Weekly Weekly
30 % savings
3 35 % savings
Compute
Yearly

75. How do I leverage Amazon Auto Scaling
effectively ?

76. Leave it to the experts , we will
handle this
Cloud Architecture Consulting
Cloud Application Development
Cloud Migration & Implementation
Cloud Adoption Strategy
“Let's get the job done”

77. Contact Us
“All you need is an idea and the cloud will execute it for you.” (Structure 2010 event)
- Dr Werner Vogels , CTO of Amazon
For more details on how we can help your business , contact
Harish11g.aws@gmail.com
http://harish11g.blogspot.com
www.twitter.com/harish11g
http://www.linkedin.com/in/harishganesan