Change EC2 instance type from Intel to AMD

Opportunity Name:

Change EC2 instance type from Intel to AMD

 

AWS Resource Type:

EC2 Instance

 

Opportunity Description:

The AMD-based EC2 instances that AWS offers are typically about 10% cheaper than the equivalent Intel EC2 instances but provide a very similar level of performance. This FF identifies Intel EC2 instances that can be safely retyped to their AMD counterparts.

 

Specifically, the instance families that this FF will retype are:

T3 -> T3a, M5 -> M5a, C5 -> C5a, R5 -> R5a, M6i -> M6a, C6i -> C6a, R6i -> R6a, and M7i -> M7a.

 

The FF requires a short downtime for each instance that is retyped because the instance must be restarted for the change to take effect.

 

Criteria for identifying the opportunity:

This FF considers the following EC2 instances:

•  Only Intel instance family types deemed safe to change to their AMD counterpart: T3, C5, M5, R5, C6i, M6i, R6i, and M7i.
•  Instances for which we do not have at least 30 days of CPU utilization data (roughly 70% coverage of that period) are excluded.
•  Instances where the maximum CPU utilization over the last 30 days was 90% or higher are excluded. This 90% ceiling (a 10% headroom buffer, to account for the AMD counterpart's slightly lower clock frequency) applies uniformly across all of the families above — it is not a different threshold per family.
•  EBS throughput/IOPS and network throughput are measured and reported, but they do not gate the recommendation: AMD instances are always equal to or better than their Intel counterpart of the same size for EBS and network performance. Where the AMD instance's specifications for that size happen to be lower than the Intel instance's, CloudFix automatically monitors for a performance regression after the migration and can roll it back (see below).
•  Instance types that are not long-lived are excluded — for example, instances that are part of an Auto Scaling Group, an Amazon EMR cluster, or an Amazon EKS cluster.
•  AWS Marketplace/DevPay instances are excluded.
•  The instance's annualized cost, based on the last ~30 days of usage, must exceed the tenant's cost threshold, which defaults to $100/year and is configurable.
•  Instances covered by a Reserved Instance or Savings Plan are not excluded — the cost used to identify and size the opportunity already reflects that amortized, discounted price.

 

Potential savings (range in % on an annual basis):

Across the CloudFix user base, we have found that roughly 25% of EC2 spending is on-demand instances using Intel types that can be migrated to AMD. We can reduce around 10% of its costs.

 

What happens when the Fixer is executed?

The fixer does the following:

•  Take an EBS snapshot while the instance is running (skipped if a recent snapshot already exists).
•  Stop the EC2 instance.
•  Take a second EBS snapshot now that the instance is stopped.
•  Change the instance type to its AMD counterpart.
•  Start the EC2 instance.
•  If it started successfully, the fixer is done. Where the AMD instance's EBS (IOPS/throughput) or network specifications for that size are lower than the original Intel instance's, CloudFix also sets up CloudWatch alarms for those specific metrics on that instance, which allow an automatic rollback if performance regresses.
•  If the instance failed to start, initiate an automatic rollback which changes the type back to Intel and attempts to start the instance.
•  If it still fails to start, CloudFix reports the error (there is no automated runbook for this step); the customer needs to repair and restart the instance (if necessary, the EBS snapshot can be used to create a new EC2 instance). This scenario is highly unlikely.

 

Is it possible to rollback once CloudFix implements the fixer?

Yes. If the instance fails to start after being retyped, CloudFix automatically changes the instance type back to the original Intel type and attempts to start it. In the rare case that this automatic rollback also fails, CloudFix reports the error so the customer can intervene manually, using the EBS snapshot to build a new EC2 instance if necessary.

 

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

Yes.

 

Does this fix require downtime?

Yes. The EC2 instance is rebooted.

 

Additional Resources:

• CloudFix Knowledge Base
• AWS EC2 Instance Types
• Stopping and Starting Your Instances