Which scaling policy type should I use?

Target tracking is the simplest and works well for most workloads — you set a target (e.g., 50% CPU) and Auto Scaling handles the math. Step scaling gives you more control with discrete scaling actions at different alarm thresholds, useful when you need asymmetric scale-out vs scale-in behavior. Predictive scaling analyzes 14 days of historical data to forecast future demand and schedule capacity changes in advance. For best results, combine predictive scaling (for known patterns) with target tracking (for unexpected demand).

What is the difference between cooldown and warm-up?

Cooldown is a period after a scaling action during which Auto Scaling ignores further alarm triggers, preventing rapid oscillation. Warm-up (instance warm-up) tells Auto Scaling how long a new instance takes to reach steady state — during warm-up, the instance is not counted toward aggregate metrics. Setting warm-up too low means metrics spike down as unready instances are counted, triggering unnecessary scale-out. Setting it too high delays metric stabilization. Typical warm-up values range from 60 to 300 seconds depending on application startup time.

Auto Scaling Policy Builder

ComputeAWS

Build Auto Scaling target tracking, step scaling, and predictive scaling policies.

Last verified: May 2026

Auto Scaling Configuration

Build Auto Scaling target tracking, step scaling, and predictive scaling policies.

Required Fields

AutoScalingGroupNamePolicyNamePolicyType

{
  "AutoScalingGroupName": "prod-web-asg",
  "PolicyName": "cpu-target-tracking",
  "PolicyType": "TargetTrackingScaling",
  "TargetTrackingConfiguration": {
    "PredefinedMetricSpecification": {
      "PredefinedMetricType": "ASGAverageCPUUtilization"
    },
    "TargetValue": 60.0,
    "ScaleInCooldown": 300,
    "ScaleOutCooldown": 60,
    "DisableScaleIn": false
  },
  "StepScalingPolicies": [
    {
      "PolicyName": "scale-out-high-cpu",
      "PolicyType": "StepScaling",
      "AdjustmentType": "ChangeInCapacity",
      "StepAdjustments": [
        {
          "MetricIntervalLowerBound": 0,
          "MetricIntervalUpperBound": 20,
          "ScalingAdjustment": 1
        },
        {
          "MetricIntervalLowerBound": 20,
          "ScalingAdjustment": 3
        }
      ],
      "Cooldown": 120,
      "MetricAggregationType": "Average"
    }
  ],
  "PredictiveScaling": {
    "PolicyName": "predictive-cpu",
    "MetricSpecifications": [
      {
        "TargetValue": 50,
        "PredefinedMetricPairSpecification": {
          "PredefinedMetricType": "ASGCPUUtilization"
        }
      }
    ],
    "Mode": "ForecastAndScale",
    "SchedulingBufferTime": 300
  }
}

Generated Output

Output will appear here...

About This Tool

Auto Scaling policies determine how and when EC2 Auto Scaling groups add or remove instances in response to demand. AWS offers three policy types: target tracking (maintain a metric at a target value), step scaling (scale in discrete steps based on alarm thresholds), and predictive scaling (use machine learning to forecast demand and pre-scale). Choosing the right policy type and configuring the scaling parameters correctly is critical for balancing performance with cost. The Auto Scaling Policy Builder helps you configure all three policy types with proper metric selections, cooldown periods, and scaling adjustments.

Real-World Scenario

Your team's 30-instance Auto Scaling group has been oscillating: scale-out to 60 instances during morning peak, then thrashing as instances finish boot just as load drops. The builder helps configure: target tracking on average CPU at 60% target, instance warm-up of 180 seconds (your Java app's real ready time), plus predictive scaling for the known 9am peak. After deploy, scaling becomes smooth — no more 60-instance overshoots, no more stampedes. Average instance count drops from 38 to 24, saving ~$3K/month while improving response time consistency.

When to Use This Tool

•Setting up target tracking policies that maintain average CPU utilization at 60% across an Auto Scaling group
•Configuring step scaling policies with multiple thresholds for aggressive scale-out and gradual scale-in
•Building predictive scaling policies that pre-warm capacity before recurring daily traffic spikes
•Combining target tracking with predictive scaling for applications with both predictable and unpredictable traffic patterns

Pro Tips

TIP

Target tracking is the right default for 95% of workloads. The other policy types feel more powerful but add complexity that's rarely justified. Step scaling is genuinely useful for asymmetric scaling (aggressive scale-out, gradual scale-in to avoid hammering downstream); predictive scaling needs at least 14 days of consistent traffic patterns to work well.

TIP

Instance warm-up time MUST match your application's actual time to start serving traffic at full capacity (not just `systemctl start`). For Java apps with JIT compilation, that's often 90-180 seconds. Setting it too low triggers cascading scale-outs; too high causes capacity gaps during real spikes.

TIP

Predictive scaling pre-warms capacity 15 minutes BEFORE a forecasted demand peak. For traffic that ramps up sharply at a known time (9am login storm, daily report runs), this eliminates the 'always 5 minutes behind' pain of reactive scaling. Combine with target tracking as a safety net for unforecasted spikes.

How It Works Under the Hood

The builder generates Auto Scaling policies for each type: target tracking (specify metric type, target value, optional disable scale-in), step scaling (alarm thresholds with scaling adjustments), or predictive scaling (metric specification, mode: ForecastAndScale or ForecastOnly, scheduling buffer time). Output is generated as aws autoscaling put-scaling-policy commands and Terraform aws_autoscaling_policy resources, plus the underlying CloudWatch alarms for step scaling.

Frequently Asked Questions

Which scaling policy type should I use?: Target tracking is the simplest and works well for most workloads — you set a target (e.g., 50% CPU) and Auto Scaling handles the math. Step scaling gives you more control with discrete scaling actions at different alarm thresholds, useful when you need asymmetric scale-out vs scale-in behavior. Predictive scaling analyzes 14 days of historical data to forecast future demand and schedule capacity changes in advance. For best results, combine predictive scaling (for known patterns) with target tracking (for unexpected demand).
What is the difference between cooldown and warm-up?: Cooldown is a period after a scaling action during which Auto Scaling ignores further alarm triggers, preventing rapid oscillation. Warm-up (instance warm-up) tells Auto Scaling how long a new instance takes to reach steady state — during warm-up, the instance is not counted toward aggregate metrics. Setting warm-up too low means metrics spike down as unready instances are counted, triggering unnecessary scale-out. Setting it too high delays metric stabilization. Typical warm-up values range from 60 to 300 seconds depending on application startup time.

Related Learning Guides

EC2 Instance Types Explained24 min read

Was this tool helpful?

Disclaimer: This tool runs entirely in your browser. No data is sent to our servers. Always verify outputs before using them in production. AWS, Azure, and GCP are trademarks of their respective owners.

Auto Scaling Policy Builder

ComputeAWS

Build Auto Scaling target tracking, step scaling, and predictive scaling policies.

Last verified: May 2026

Auto Scaling Configuration

Build Auto Scaling target tracking, step scaling, and predictive scaling policies.

Required Fields

AutoScalingGroupNamePolicyNamePolicyType

Generated Output

Output will appear here...

About This Tool

Real-World Scenario

When to Use This Tool

•Setting up target tracking policies that maintain average CPU utilization at 60% across an Auto Scaling group
•Configuring step scaling policies with multiple thresholds for aggressive scale-out and gradual scale-in
•Building predictive scaling policies that pre-warm capacity before recurring daily traffic spikes
•Combining target tracking with predictive scaling for applications with both predictable and unpredictable traffic patterns

Pro Tips

TIP

How It Works Under the Hood

Frequently Asked Questions

Which scaling policy type should I use?: Target tracking is the simplest and works well for most workloads — you set a target (e.g., 50% CPU) and Auto Scaling handles the math. Step scaling gives you more control with discrete scaling actions at different alarm thresholds, useful when you need asymmetric scale-out vs scale-in behavior. Predictive scaling analyzes 14 days of historical data to forecast future demand and schedule capacity changes in advance. For best results, combine predictive scaling (for known patterns) with target tracking (for unexpected demand).
What is the difference between cooldown and warm-up?: Cooldown is a period after a scaling action during which Auto Scaling ignores further alarm triggers, preventing rapid oscillation. Warm-up (instance warm-up) tells Auto Scaling how long a new instance takes to reach steady state — during warm-up, the instance is not counted toward aggregate metrics. Setting warm-up too low means metrics spike down as unready instances are counted, triggering unnecessary scale-out. Setting it too high delays metric stabilization. Typical warm-up values range from 60 to 300 seconds depending on application startup time.

Related Learning Guides

EC2 Instance Types Explained24 min read

Was this tool helpful?