Auto Scaling

Icon

AutoScaling Icon

Problem

• Most application during its life-time receives varying load.

• To ensure that application is highly available, it is necessary to make sure that the application has the resources to handle the load.

  • Meaning, application can scale-out to match the increasing load and scale-in when load is decreased.

• Remove instances which are unhealthy.

• Ensure minimum/maximum application instances are running at a given time.

• Automatically register new instances to the load balancer.

Solution

• Automatic Scaling Group (ASG) is the solution to above set of problems.

• ASG is linked to a load balancer.

• ASG in itself doesnt solve the problem of scalability. ELB and ASG and multi AZ together help us achieve Scalability and High Availability.

• ASG is free.

Configuring ASG

• A Launch Template is used to configure the ASG.

• Parameters include,

  • AMI

  • Instance type

  • EC2 User Data

  • EBS/EFS volume

  • SSH key pair

  • Security group

  • IAM Roles

  • ELB configuration

  • Network configuration

• Following scaling parameters are to be configured in a Launch Template as part of scaling policies when creating an ASG.

  • Desired capacity: Represents the initial capacity of the Auto Scaling group at the time of creation.

  • Minimum capacity: an Auto Scaling group cannot decrease its desired capacity lower than the minimum size limit.

  • Maximum capacity: Represents the maximum group size. Desired capacity cannot exceed this capacity value.

Scaling policy

• Policies based on which scaling can be done.

Dynamic Scaling

Target Tracking Scaling

• Most simple and easy to set-up.

• For example, based on Average CPU utilization configuration.

Simple Scaling

• Based on cloud watch alarm, one can setup scale-out and scale-in configuration.

Step Scaling

• Same as simple scaling, but scaling can be done in steps with out scaling cool down period between scaling activity.

Scheduled Actions

• Anticipate scaling based on known usage pattern.

Predective Scaling

• Based on historical load, forecast load and scheduled scaling ahead. This will scale-in or scale-out the instances.

• This uses Machine learning.

Scaling Metrics

• Average CPU Utilization

• Request Count per target

• Average Network In/Out

• Custom metrics configured using Cloud Watch

Scaling cooldown

• After a scaling activity, for a duration of cooldown period (by default 5 minutes, though configurable) the ASG will not launch or terminate additional instances.

• During this period ASG will wait for the metrics to stabilize.

Instance Refresh

• Instance refresh helps us to update launch template configuration based on which new instances should be created using this new template.

• Updating launch template should lead to terminating instances running on old template and start new instances using newer template.

• This feature enables us to keep an healthy percentage of EC2 instances. This percentage help us to keep a track of how many instances using old template can be deleted at a time.

• One can specify minimum healthy percentage and warmup time after which only the newly set up instances will be considered ready to handle traffic.

References

• Difference between Simple scaling and Step scaling