Optimize EC2 Auto Scaling Group costs

Opportunity Name:

Optimize EC2 Auto Scaling Group costs

AWS Resource Type:

EC2

Opportunity Description:

While Auto Scaling Groups (ASGs) themselves do not incur costs, the EC2 instances that they utilize are typically the largest contributor to an organization's AWS costs and these EC2 instances are often overprovisioned. This FF reduces ASG costs by using attribute-based instance type selection to allow AWS to select the lowest-cost instance types that match the required instance attributes (CPU, memory, networking).

Criteria for identifying the opportunity:

This FF identifies ASGs that meet the following criteria:

• The ASG is not externally managed by CloudFormation, AWS Migration Hub, EMR, or EKS.
• The ASG is not managed by IaC (e.g. Terraform - this is a heuristic based on resource tags)
• The ASG is managing ephemeral instances (last instance launch time is less than some threshold, e.g. 3 months)
• The ASG is not using a Warm Pool
• The ASG uses a Launch Template and the Launch Template does not have Overrides.
• The ASG does not already use attribute-based instance type selection.

Potential savings (range in % on annual basis):

The actual savings will depend on how your organization uses ASGs, but we estimate that this FF can realize a ~5% cost reduction for each ASG that this fixer is applied to.

What happens when the Fixer is executed?

• The Fixer creates a new Launch Template for the ASG that matches the existing one, except for the following changes:
  • Use the attribute-based instance type selection requirements that were identified by the Finder.
  • Use the lowest-price allocation strategy.
  • Only allow Intel to/from AMD CPU architecture changes.
  • Set the price protection thresholds for On-Demand instances and Spot instances to 0.
  • Temporarily increases the desired capacity by 1 (and the max size if needed) so that it can test the launch template immediately
• The Fixer applies the new launch template to the ASG (keeping the old launch template intact in case a rollback is needed)
• The Fixer checks that the desired capacity increase is fulfilled successfully by a new healthy instance
  • If it wasn’t fulfilled successfully, then rollback by re-applying the original launch template
• Either way decrease the desired capacity back to the original number (and the max size if needed)
• Monitor the GroupPendingCapacity CloudWatch metric - if it is greater than zero for 30 minutes initiate a rollback (i.e. re-apply to the ASG the original launch template)
• After at least 30 days of monitoring without issues and assuming at least one instance was launched successfully, cleanup the old launch template and remove the CloudWatch alarm

Is it possible to rollback once CloudFix implements the fixer?

Yes - CloudFix will automatically rollback the changes if the ASG’s GroupPendingCapacity CloudWatch metric is greater than zero for 30 minutes or more.

Can CloudFix implement the fix automatically once I accept the recommendation?

Yes.

Does this fix require downtime?

No - only newly launched instances will use the new launch template instance configuration. Existing instances will continue to run until terminated by the ASG.

 

Additional Resources:

• CloudFix Knowledge Base
• Attribute-based instance type selection