Elastic Scaling with AWS Autoscaling groups

• Lets do the elastic scaling based on CPU Utilization
• My website works well when CPU Utilization is less than 80%
• Lets start our application with 1 ec2 instance (desired capacity)
• Lets also define the max and min instances
  • min: 1
  • max: 5
• AWS offers us three ways
  • Target Tracking Policy
  • Simple Scaling
  • Step Scaling
• For now lets use Target Tracking Policy: In this we define the ideal CPU for your application.
• Note for artificial load we will use the stress tool installed on the machine stress --cpu 8 --io 4 --vm 2 --vm-bytes 128M -v --timeout 100m
• For screenshots refer classroom video
• Simple Scaling:
  • Works with Cloudwatch alarm
  • When an alarm is triggered scale out or in happens depending on configuration
  • Cool down period: This is the time after scaling where no other scaling will be allowed
• Step Scaling:
  • Similar to simple scaling but no cooldown periods
• AWS also supports Predictive Scaling

Target Tracking Scaling Policy vs. Simple & Step Scaling in AWS Auto Scaling

---

AWS Auto Scaling supports three types of scaling policies:
1. Simple Scaling – Adds/removes a fixed number of instances per alarm trigger.
2. Step Scaling – Adjusts scaling in steps based on how far the metric deviates from thresholds.
3. Target Tracking Scaling – Automatically adjusts capacity to maintain a target metric value.

---

What is Target Tracking Scaling Policy?

• Works like a thermostat: Automatically increases or decreases the number of instances to maintain a target metric value.
• AWS continuously monitors the metric and scales dynamically.
• No need to define specific step actions; AWS handles everything.

How It Works

1. Define a target metric (e.g., CPU utilization at 50%).
2. AWS Auto Scaling adjusts instances dynamically to maintain this target.
3. If the metric exceeds the target, AWS adds instances.
4. If the metric falls below the target, AWS removes instances.
5. AWS handles all calculations, so no manual step configuration is needed.

---

Key Differences Between Simple, Step, and Target Tracking Scaling

Feature	Simple Scaling	Step Scaling	Target Tracking Scaling
Scaling Trigger	Fixed threshold	Step-based thresholds	Target-based metric
Scaling Adjustment	Adds/removes a fixed number of instances per alarm	Adds/removes instances based on how far the metric deviates	AWS automatically adjusts to maintain the target
Response Speed	Slower (cooldown period required)	Faster (no cooldown)	Fastest (reacts dynamically)
Configuration Complexity	Simple but rigid	More control, requires tuning	Easiest (AWS auto-manages)
Example Use Case	Predictable, slow-changing loads	Large, unpredictable spikes	Maintaining steady performance (e.g., CPU at 50%)

---

Example Use Cases for Each Policy

1. Simple Scaling – Use when you need basic auto scaling with a fixed instance count change (e.g., “always add 1 instance if CPU > 80%”).
2. Step Scaling – Use when different levels of scaling are required based on how much load increases (e.g., “add 1 instance at 70%, 2 at 85%, 3 at 95%”).
3. Target Tracking – Use when you want automatic scaling without manual thresholds (e.g., “keep CPU usage at 50% at all times”).

---

Example: Setting Up Target Tracking in AWS Auto Scaling

Step 1: Define Target Metric

• Go to EC2 Auto Scaling Groups → Select your Auto Scaling Group.
• Click Automatic Scaling → Add a policy.
• Select Target Tracking Scaling.

Step 2: Set Scaling Target

• Choose a metric (e.g., CPU Utilization).
• Set the target value (e.g., 50% CPU).
• AWS automatically adjusts instances to keep CPU close to 50%.

Step 3: Monitor Scaling

• AWS dynamically adds/removes instances to maintain CPU at 50%.
• No manual step configuration is needed.

---

Final Thoughts: Which One to Use?

Use Case	Recommended Policy
Basic, predictable scaling	Simple Scaling
Large, sudden traffic spikes	Step Scaling
Fully automated, self-adjusting scaling	Target Tracking Scaling

If you don’t want to manually define thresholds and steps, Target Tracking Scaling is the best choice.