#aws #asg #scaling #policies #tracking

Scaling policies

  • Target tracking: It based on average ASG CPU like around 40%
  • Simple/Step Tracking: If CPU is higher than 70%, add 2 units, if <30%, reduce 1 CPU unit
  • Schedule Actions: Scale based on time.
  • Predictive Scaling: Using ML behind the scene and periodic what time that needs to be scaled and based on some metrics.

Some metrics for scaling:

  • CPU
  • RequestPerTarget
  • Network in/out
  • etc, …

Specifications:

  • There are default cooldown for ASG to stop spawn/terminate instance to get the metrics to be stabilized.
  • We can reduce the CD by using ready-to-use AMI for EC2 instance, so the metric will be stabilized as soon as possible.