Create mixed instances group using attribute-based instance type selection

Instead of manually choosing instance types for your mixed instances group, you can specify a set of instance attributes that describe your compute requirements. As Amazon EC2 Auto Scaling launches instances, any instance types used by the Auto Scaling group must match your required instance attributes. This is known as attribute-based instance type selection.

This approach is ideal for workloads and frameworks that can be flexible about which instance types they use, such as containers, big data, and CI/CD.

The following are benefits of attribute-based instance type selection:

How attribute-based instance type selection works

With attribute-based instance type selection, instead of providing a list of specific instance types, you provide a list of instance attributes that your instances require, such as:

There are many options available for defining your instance requirements. For a description of each option and the default values, see InstanceRequirements in the Amazon EC2 Auto Scaling API Reference.

When your Auto Scaling group needs to launch an instance, it will search for instance types that match your specified attributes and are available in that Availability Zone. The allocation strategy then determines which of the matching instance types to launch. By default, attribute-based instance type selection has a price protection feature enabled to prevent your Auto Scaling group from launching instance types that exceed your budget thresholds.

By default, you use the number of instances as the unit of measurement when setting the desired capacity of your Auto Scaling group, meaning each instance counts as one unit.

Alternatively, you can set the value for desired capacity to the number of vCPUs or the amount of memory. To do so, use the Desired capacity type dropdown field in the AWS Management Console or the DesiredCapacityType property in the CreateAutoScalingGroup or UpdateAutoScalingGroup API operation. Amazon EC2 Auto Scaling then launches the number of instances required to meet the desired vCPU or memory capacity. For example, if you use vCPUs as the desired capacity type and use instances with 2 vCPUs each, a desired capacity of 10 vCPUs would launch 5 instances. This is a useful alternative to instance weights.

Price protection

With price protection, you can specify the maximum price you are willing to pay for EC2 instances launched by your Auto Scaling group. Price protection is a feature that prevents your Auto Scaling group from using instance types that you would consider too expensive even if they happen to fit the attributes that you specified.

Price protection is enabled by default and has separate price thresholds for On-Demand Instances and Spot Instances. When Amazon EC2 Auto Scaling needs to launch new instances, any instance types priced above the relevant threshold are not launched.

On-Demand price protection

For On-Demand Instances, you define the maximum On-Demand price you're willing to pay as a percentage higher than an identified On-Demand price. The identified On-Demand price is the price of the lowest priced current generation C, M, or R instance type with your specified attributes.

If an On-Demand price protection value is not explicitly defined, a default maximum On-Demand price of 20 percent higher than the identified On-Demand price will be used.

Spot price protection

By default, Amazon EC2 Auto Scaling will automatically apply optimal Spot Instance price protection to consistently select from a wide range of instance types. You can also manually set the price protection yourself. However, letting Amazon EC2 Auto Scaling do it for you can improve the likelihood that your Spot capacity is fulfilled.

You can manually specify the price protection using one of the following options. If you manually set the price protection, we recommend using the first option.

Customize price protection

You can customize the price protection thresholds in the Amazon EC2 Auto Scaling console or using the AWS CLI or SDKs.

If you set Desired capacity type (DesiredCapacityType) to vCPUs or Memory GiB, the price protection applies based on the per vCPU or per memory price instead of the per instance price.

You can also turn off price protection. To indicate no price protection threshold, specify a high percentage value, such as 999999.

Performance protection

Performance protection is a feature that ensures your Auto Scaling group uses instance types that are similar to or exceed a specified performance baseline. To use performance protection, you specify an instance family as a baseline reference. The capabilities of the specified instance family establish the lowest acceptable level of performance. When Auto Scaling selects instance types, it considers your specified attributes and the performance baseline. Instance types that fall below the performance baseline are automatically excluded from selection, even if they match your other specified attributes. This ensures that all selected instance types offer performance similar to or better than the baseline established by the specified instance family. Auto Scaling uses this baseline to guide instance type selection, but there is no guarantee that the selected instance types will always exceed the baseline for every application.

Currently, this feature only supports CPU performance as a baseline performance factor. The CPU performance of the specified instance family serves as the performance baseline, ensuring that selected instance types are similar to or exceed this baseline. Instance families with the same CPU processors lead to the same filtering results, even if their network or disk performance differs. For example, specifying either c6in or c6i as the baseline reference would produce identical performance-based filtering results because both instance families use the same CPU processor.

Unsupported instance families

The following instance families are not supported for performance protection:

If you enable performance protection by specifying a supported instance family, the returned instance types will exclude the above unsupported instance families.

Example: Set a CPU performance baseline

In the following example, the instance requirement is to launch with instance types that have CPU cores that are as performant as the c6i instance family. This will filter out instance types with less performant CPU processors, even if they meet your other specified instance requirements such as the number of vCPUs. For example, if your specified instance attributes include 4 vCPUs and 16 GB of memory, an instance type with these attributes but with lower CPU performance than c6i will be excluded from selection.

"BaselinePerformanceFactors": {
        "Cpu": {
            "References": [
                {
                    "InstanceFamily": "c6i"
                }
            ]
        }

Considerations

Consider the following when using performance protection:

Prerequisites

Create a mixed instances group with attribute-based instance type selection (console)

Use the following procedure to create a mixed instances group by using attribute-based instance type selection. To help you move through the steps efficiently, some optional sections are skipped.

For most general purpose workloads, it's enough to specify the number of vCPUs and memory that you need. For advanced use cases, you can specify attributes like storage type, network interfaces, CPU manufacturer, and accelerator type.

To review the best practices for a mixed instances group, see Setup overview for creating a mixed instances group.

Create a mixed instances group with attribute-based instance type selection (AWS CLI)

To create a mixed instances group using the command line

Use one of the following commands:

Example configuration

To create an Auto Scaling group with attribute-based instance type selection by using the AWS CLI, use the following create-auto-scaling-group command.

The following instance attributes are specified:

aws autoscaling create-auto-scaling-group --cli-input-json file://~/config.json

{
    "AutoScalingGroupName": "my-asg",
    "DesiredCapacityType": "units",
    "MixedInstancesPolicy": {
        "LaunchTemplate": {
            "LaunchTemplateSpecification": {
                "LaunchTemplateName": "my-launch-template",
                "Version": "$Default"
            },
            "Overrides": [{
                "InstanceRequirements": {
                    "VCpuCount": {"Min": 4, "Max": 8},
                    "MemoryMiB": {"Min": 16384},
                    "CpuManufacturers": ["intel"]
                }
            }]
        },
        "InstancesDistribution": {
            "OnDemandPercentageAboveBaseCapacity": 50,
            "SpotAllocationStrategy": "price-capacity-optimized"
        }
    },
    "MinSize": 0,
    "MaxSize": 100,
    "DesiredCapacity": 4,
    "DesiredCapacityType": "units",
    "VPCZoneIdentifier": "subnet-5ea0c127,subnet-6194ea3b,subnet-c934b782"
}

aws autoscaling create-auto-scaling-group --cli-input-yaml file://~/config.yaml

---
AutoScalingGroupName: my-asg
DesiredCapacityType: units
MixedInstancesPolicy:
  LaunchTemplate:
    LaunchTemplateSpecification:
      LaunchTemplateName: my-launch-template
      Version: $Default
    Overrides:
    - InstanceRequirements:
         VCpuCount:
           Min: 2
           Max: 4
         MemoryMiB:
           Min: 2048
         CpuManufacturers:
         - intel
  InstancesDistribution:
    OnDemandPercentageAboveBaseCapacity: 50
    SpotAllocationStrategy: price-capacity-optimized
MinSize: 0
MaxSize: 100
DesiredCapacity: 4
DesiredCapacityType: units
VPCZoneIdentifier: subnet-5ea0c127,subnet-6194ea3b,subnet-c934b782

Preview your instance types

You can preview the instance types that match your compute requirements without launching them and adjust your requirements if necessary. When creating your Auto Scaling group in the Amazon EC2 Auto Scaling console, a preview of the instance types appears in the Preview matching instance types section on the Choose instance launch options page.

Alternatively, you can preview the instance types by making an Amazon EC2 GetInstanceTypesFromInstanceRequirements API call using the AWS CLI or an SDK. Pass the InstanceRequirements parameters in the request in the exact format that you would use to create or update an Auto Scaling group. For more information, see Preview instance types with specified attributes in the Amazon EC2 User Guide.

Related resources

To learn more about attribute-based instance type selection, see Attribute-Based Instance Type Selection for EC2 Auto Scaling and EC2 Fleet on the AWS Blog.

You can declare attribute-based instance type selection when you create an Auto Scaling group using AWS CloudFormation. For more information, see the example snippet in the Auto scaling template snippets section of the AWS CloudFormation User Guide.