Use performance metrics as capacity units

Key concepts

Scaling rules

• A scale-out event is triggered when the total capacity is less than the desired capacity or the minimum capacity.
• A scale-in event is triggered when the total capacity is greater than or equal to the sum of the desired capacity and the largest weight.

Usage notes

• You must set a weight for every instance type in the scaling group.
• If you remove an instance type from the scaling configuration, the weight of existing instances of that type within the scaling group remains unchanged.
• When you modify the weight of an instance type with existing instances, Auto Scaling immediately recalculates the group's current capacity. This may trigger a new scaling activity.

Procedure

Use this procedure to set instance type weights when a scaling configuration is the configuration source for the scaling group.

1. Create a scaling group.
   This step focuses on options related to the multi-zone scaling policy. For other scaling group options, see Configure a scaling group.
   1. Set VPC to VPC and then select multiple vSwitches within that VPC.
      Each vSwitch belongs to a single zone. Selecting multiple vSwitches lets you create ECS instances across zones to make better use of available resources.
   2. Set the Cost Optimization Policy policy to Cost Optimized.
   3. Configure the remaining scaling group options as needed.
2. Create a scaling configuration.
   This step focuses on setting the instance type weight based on vCPU count. For other scaling configuration options, see Create a scaling configuration for ECS instances.
   1. Set the Pay-as-you-go to pay-as-you-go.
   2. Select multiple instance types, up to a maximum of 10.
   3. Select Set vCPU Capacity. The system automatically sets a weight for each selected instance type based on its vCPU count.
      For example, for the selected instance types ecs.c5.large (2 vCPUs, 4 GiB), ecs.c5.xlarge (4 vCPUs, 8 GiB), and ecs.c5.2xlarge (8 vCPUs, 16 GiB), the weights are automatically set to 2, 4, and 8, respectively.

      You can also customize the weights. When you customize weights, follow these recommendations:

      • Use a performance metric that is relevant to your workload, such as the number of vCPUs or the amount of memory in GiB. You can use a single vCPU, 1 GiB of memory, or your least powerful instance type as one capacity unit. The capacity of the scaling group is then calculated relative to this unit.
      • Set appropriate weight values. Ensure that the current capacity of the scaling group is two to three times the largest weight defined for an instance type.
      • Avoid setting vastly different weights for different instance types. For example, do not set the weight of a small instance type to 1 and a large instance type to 200. A large gap between weights can negatively affect the cost-effectiveness of the scaling group.

      For information about the priority of multiple instance types when instances are created, see How the weighted unit price is calculated.

   4. Configure the remaining scaling configuration options as needed.
3. Enable the scaling group.
4. Create a scaling rule.
   This step describes the options required for a simple validation test. For other scaling rule options, see Configure scaling rules.
   1. Set the Simple Scaling Rule to Simple Scaling Rule.
   2. Set the Action to Add 10 capacity units.
   3. Configure the remaining scaling rule options as needed.
5. Execute the scaling rule.
   In this example, the ecs.c5.2xlarge instance type has the lowest weighted unit price. Therefore, the scaling group launches two ecs.c5.2xlarge instances, adding 16 capacity units to the scaling group.

Weighted unit price calculation

If your scaling group uses the Cost Optimization Policy and you have set instance weights, Auto Scaling launches the instance types with the lowest weighted unit price during scale-out. For more information, see Reduce costs by using a scaling policy and multiple instance types.